Professional Documents
Culture Documents
TH2018PESC1030
TH2018PESC1030
par
I would like to express my deep gratitude to my research supervisors, for their patient
guidance, enthusiastic encouragement and useful critiques of this research work.
I wish to acknowledge the support and also the guidance received from personal of
Géodynamique & Structure. Thanks for your trust in the project. Worth to mention the
support received by the "Consejo Nacional de Ciencia y Tecnología (CONACYT)" and the
"Consejo Mexiquense de Ciencia y Tecnología (COMECyT)". In particular, I am grateful
for the opportunity they gave me to share my project and experiences in my country.
I would also like to extend my thanks to the technicians, the interns, the students and
the staff of the ESTP: thank you for the time we shared.
I wish to thank those who were always there; my friends, my girlfriend, my family, and
specially quiero agradecer a mis padres.
Contents
Acknowledgements ii
Contents vi
Introduction 2
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Summary of research contributions . . . . . . . . . . . . . . . . . . . . . . . . . 7
Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 Comparative example 98
4.1 Example selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.2 2D planar model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3 Surface model representation . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.4 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.5 3D brick model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.7.1 Capabilities of the algorithm . . . . . . . . . . . . . . . . . . . . . . 118
4.7.2 Quantity of reinforcement . . . . . . . . . . . . . . . . . . . . . . . 120
Bibliography 133
Within the field of Reinforced Concrete (RC) structures and more specifically, at the design
of non-flexural elements such as corbels, nibs, and deep beams, the rational procedure of
conception and justification referred as Strut-and-Tie Method (STM) has shown some
advantages over classical algorithms of reinforcement computation based on FE analysis
(eg. Wood-Armer or Capra-Maury).
The STM remains a suitable alternative for the design of concrete structures presenting
either elastic or plastic behaviour whose application framework is well defined in concrete
structures’ design codes like the EuroCodes and the AASHTO-LRFD Bridge Design Spec-
ifications. Nevertheless, this method has the main inconvenient of requiring a high amount
of resources investment in terms of highly experienced personal or in terms of computa-
tional capacity for, respectively, its manual application or an automatic approach through
topology optimisation.
The document proposes a light alternative, in terms of required iterations, to the au-
tomation of the STM, which starts from the statement that the resultant struts and ties of
a suitable ST model can be distributed according to the direction of the principal stresses,
σIII and σI , obtained from a planar or a three-dimensional FE model.
Résumé
Dans le domaine des structures en Béton Armé (BA) et plus spécifiquement, lors de la con-
ception d’éléments non-flexibles tels que les corbeaux, les poutres bayonnetts et les poutres
profondes, la Méthode Bielle-Tirant (MBT) présente des avantages par rapport aux algo-
rithmes classiques de calcul de ferraillage basé sur l’analyse FE (par exemple Wood-Armor
ou Capra-Maury).
La Methode Bielle-Tirant reste une alternative adaptée pour la conception de structures
en béton présentant un comportement élastique ou plastique dont le cadre d’application est
bien défini dans les codes de conception des structures en béton comme les EuroCodes et les
spécifications de conception des ponts AASHTO-LRFD. Néanmoins, cette méthode présente
l’inconvénient majeur de nécessiter un investissement important en ressources humaines
ou en capacité de calcul pour, respectivement, son application manuelle ou une approche
automatique par optimisation de topologie.
Le document propose une alternative légère, en termes d’itérations requises, à l’automa-
tisation de la MBT, qui part de l’affirmation que les entretoises résultantes et les attaches
d’un modèle ST approprié peuvent être distribuées selon la direction des contraintes prin-
cipales, σIII et σI , obtenus à partir d’un modèle préliminaire aux EF.
3
Background
In the industrial context of engineering, most of the structural elements in reinforced con-
crete structures are conceived under the Bernoulli and Navier’s hypotheses. Nevertheless,
these solutions face problems when treating local non-flexural regions of the structure
where the strain distribution is significantly nonlinear (corbels, openings, gussets or near
the surroundings of concentrated loads). Sharp discontinuities can occur in the direction
of internal forces and it is imperative to provide a proper reinforcement able resist the ten-
sion while being consistent with the pertinent codes in terms of the quantity, distribution,
anchorage length, etc. In many cases, standard guides found in the construction manuals
like the ACI detailing recommendations or in the EuroCodes fulfil the design but it is up
to the structural engineer to decide whether special considerations are or are not needed.
For the exceptional cases, the structural engineer should isolate specific regions, analyse
them and provide the required reinforcement computed through a convenient methodology.
The work of Schlaich [Schlaich et al., 1987] describes in detail the use of ST models as
a reliable solution to treat the denominated D-regions (where D stands for discontinuity,
disturbance or detail). Discontinuity (which is associated with high shear stresses) is
either static (as a result of concentrated loads) or geometric (as a result of abrupt change
of geometry) or both [El-Metwally and Chen, 2017]. In brief, this procedure proposes that
the real structure should be replaced by a fictitious skeletal structure whose geometry
allows keeping the boundary conditions and load case, of the real structure, in such a
way that it complies with the Bernoulli hypothesis. This can be achieved by imposing
an idealised truss-like distribution of inner forces, where the compressive forces are taken
by inner concrete bar-type elements (struts) and the tensile forces are taken by the steel
reinforcement (ties); a third type of elements, the nodes, connect the struts and ties at their
extremities to assure the interaction between different elements. One of the shortcomings
of the applications of ST models is that, in most of the cases the model should be carried
out through a manual procedure by a highly experienced engineer.
However, the models can also be established through an elastic FE analysis of the whole
structure by considering the direction of the principal stresses along the geometry or, in
recent years, by powerful structural optimisation procedures.
Due to the interest that the STM has taken during the last decades, an effort to automa-
tise its application has been done by different research groups all around the world. Com-
puter aided approaches such as CAST [Kuchma and Tjhin, 2001], or the one established
in "Computer graphics in detailing of ST models" [Alshegeir and Ramirez, 1992], propose
tools that overlaps images of the direction fields of principal stresses over an interactive
CAD-like interface. The interface allows the user to manually propose a suitable ST model
whose distribution of elements follows the trajectory of the linear elastic stresses. This
procedure simplifies the selection and evaluation of feasible ST systems requiring, however,
the intervention of a skilled structural engineer at the earliest stage of the method.
On the other hand, structural optimisation procedures have been also applied to find
truss-like structural configurations out of plain concrete structures. Most commonly ap-
plied for aeronautics, optimisation problems can be stated for a large variety of elements
and structures found in the civil engineering field. Current approaches of structural op-
timisation are based on iterative Finite Element Analysis (FEA) where, according to its
representation, two approaches can be distinguished:
• Continuum optimisation.
• Discrete optimisation.
plies that most optimal designs have a singular matrix and present potential mechanisms
when described as a part of the ground structure leading to the second listed complication.
The third complication is related to the choice of the ground structure itself. The
ground structure approach may or may not lead to the optimal structure according to
the group of nodes proposed (quantity and position) and the set of allowed elements; the
optimal structure appears to be limited by the original geometrical restrictions and possible
connections. An alternative to overcome this difficulty is to treat the coordinates of the
structural nodes as a variable within the optimisation procedure.
Taking also the nodal coordinates as variables allows to adapt the geometry to the
boundary may induce improvements in terms of performance or weight reductions. Never-
theless, the structure is still strongly dependant on the initial layout: number of elements
and connectivity.
To summarise, regardless the fact that existent procedures are suitable for computing
ST models, the obtained trusses are still highly dependant on the selected initial truss
and, consequently, on the experience of the structural engineer. Even though the selection
of a "good" initial truss could be an easy task for "classic" or well referenced examples,
complication arises when dealing with complex load cases (in plane and out of plane loads,
multiple loads, displacements, etc.), whimsical geometries or three-dimensional models.
Ideally, a tool intended to find ST models should propose and evaluate truss-like struc-
tures keeping the experienced based decisions as minimum as possible. The results must
be feasible, not only from the point of view of mechanics but also from the point of view
of the construction codes. Additionally to these needs, the results shall be economically
viable which is an evidence that a sort of optimisation procedure must be involved.
In the context of nuclear civil works, the conception of non-flexural elements (e.g. deep
beam, joints, trimmed walls) has become such a common task that most of the times they
do not receive the detailing that they deserve during their modelling. The design and
justification of these elements is frequently based on force equilibrium along FE models
based on shell elements that must be checked for transitions between different thicknesses,
gradient in the mesh size, verification of sides-thickness ratios, etc; aspects that could
produce a diminution in the model accuracy if an exhausting detailing is not executed.
The development of a light computational tool able to threat a large variety of structural
problems and, also, able to automatically propose optimised reinforcement of concrete
structures based on the STM, could represent a huge improvement in the industrial context.
The main advantages are listed below:
• Decision support in the design process. Most of the times, decisions in the
earlier stages of design are not made based on rigorous justified structural aspects but,
they are made based on previous experience or intuition of the engineer. Computer-
aided tools can provide non-experienced engineers with a justified insight of complex
problems.
• Time saving through computer-aided design tools. Within the detailing phase,
computer-aided design has the potential to save engineer time by avoiding the need
• Prohibitive time consumption. Since design time is never unlimited, the opti-
misation procedure cannot be allowed to become critical and the amounts of time
devoted to building up a model and parameter adjusting should be small.
This document presents a consistent and rational approach for the generation of Strut-
and-Tie models for D-regions in accordance to practical reinforcement layout. The pre-
sented approach intends to present a practical alternative for the design of non-flexural
regions.
The research presented in this thesis is motivated by two principal aspects of the current
engineering practice:
2. The disparity between the vast volume of academic literature in the field of structural
optimisation and the low practical application in RC structures.
The core research objective is therefore to contribute towards reducing the evident gap
between the rational approach known as Strut-and-Tie and automatic applied method-
ologies based on finite element analysis industrially applied. The accompanying central
hypothesis is that based on the direction fields of principal stresses, plausible ground struc-
tures can be automatically proposed and optimised to create suitable Strut-an-Tie models
respecting consideration of industrial specific issues.
The research objective is achieved through the proposal of ground structure construction
technique and an investigation of optimisation methods and techniques focusing on discrete
Fully Stressed Design (FSD) for simultaneous size, topology, and geometric optimisation.
Additionally, the approach proposed within this document is coded in Matlab environ-
ment; the developed algorithm is applied and compared to studied cases of elements whose
ST models are found in the literature.
• The methods proposed in this work, discussed in chapters 3 and 4, contribute to the
development of an open computational-aided tool addressed to the building industry.
• Examples of application are presented and directly compared to results found in the
literature. Additionally, an example has been performed in order to compare results
with those expected from an industrial design.
• Following the recommendations found in the building codes (specially the Euro
Codes), an exhaustive revision of the elements has been performed not only in terms
of material resistance but also regarding spacial and size constraints.
Thesis structure
This work consists of 5 chapters intended to explore the optimisation of Strut-and-Tie
models in the structural design of non-flexural zones. Major themes are developed in this
work and an original contribution is proposed. To guide the reader, before tackling the
main subject, a brief introduction is presented at the beginning of each chapter. In the
same spirit, each chapter ends with a discussion of the content. The structure of the thesis
is then presented with an overview of each subsequent chapter.
Chapter 2 explores the vast domain of structural optimisation. This chapter summarise
the optimisation techniques and procedures applied to the optimisation of structures within
the civil works domain. The text is principally focused on the use of discrete optimisation
in the building industry, choice that is justified within the chapter. This area of application
is not intended to be exclusive, since generality is desirable in any method, but rather to
provide a unifying theme to the distinct elements of this thesis.
Chapter 3 describes and introduces the application of the developed algorithm. The
computational aided procedure is dismembered and each step and sub-algorithm is pre-
sented. In order to compare the performance of the proposed algorithm, this chapter
presents the results of its application on an example extracted from the literature. At this
stage, the comparison focuses in aesthetic aspects such as quantity of resultant elements,
distribution and inclination, geometrical aspects directly related with the optimisation pro-
cess and the automatic selection of the initial truss. A parametric study is carried out to
display the advantages and disadvantages of the application of the proposed methodology.
Chapter 4 presents a case that directly compares the results of the proposed algorithm
to those obtained through a model that follows a common engineering practice.
The document concludes by summarising the results of the preceding research and
discussing future work required in developing these methods.
Related work
The computational power of common desktop computers increases every year. This has
been one of the main aspects that have brought the use of advanced analysis programs of
research institutes closer to the designer in engineering practice. At the same time, this
has impulse the research institutes to develop new techniques and software to better suit
the needs of current practice engineers.
Listed here below are some research programs or research topics that have been devel-
oped by different research groups The list, not aiming to be exhaustive, contains notable
work that is considered to be related with the main topic of this thesis.
SPanCAD is a software for interactive design of shear walls and deep beams of irregular
geometry developed by the Technische Universiteit Delft.
The program SPanCAD is implemented on a finite element program containing only
two types of elements: a stringer element (straight bar) and a panel element (rectangle or
quadrilateral). According to [Blaauwendraad and Hoogenboom, 2002] SPanCAD is devel-
oped to apply the coarsest mesh for a given geometry.
Based on a three design step process that allows to obtain the need of reinforcement
for shear walls, deep beams and cellular structures.
The first step is the construction of the model. Using its experience and rules of thumb,
the user builds up the model by placing the stringers and panels within the structure. The
software proceeds to perform the linear-elastic analysis for all load combinations.
In the second step, the user selects the reinforcement based on force flow computed in
the precedent stage. For elements in tension the cross-section area depends on the position
and diameter of the bars. With all input quantities being determined and entered into the
program, the software performs a nonlinear analysis. The model used accounts for concrete
cracking in the tensioned stringers and panels. A revision of the steel is performed at this
point: the reinforcement of the stringers is must remain within the linear-elastic domain
while the panel reinforcement can yield.
For the third and final step, the user improves the reinforcement using the just computed
force flow and crack widths.
A strut-tie2017 http://astruttie.aroad.co.kr/index.php/advisor/
Stress field topology The stress field method has traditionally been based on the as-
sumption of a rigid-plastic stress-strain law without tensile strength for the concrete. Ne-
glecting the tensile strength of concrete requires placing a minimal amount of reinforcement
for crack control to ensure a satisfactory behaviour of the structure. This reinforcement
ensures that no brittle failure occurs at cracking and that the cracks are suitably smeared
over the element at the serviceability limit state. The development of stress fields with the
previous assumptions allows a great freedom in the choice of the load-carrying mechanism
of a structure.
Micro trusses. Nowadays, several authors intend to implement this method to the RC
field (e.g. [Zhong et al., 2016, Nagarajan et al., 2010]). The micro truss is based on the
framework method proposed by [Hrennikoff, 1941], in which the structure is replaced by an
equivalent pattern of truss elements. Then, each element is given physical characteristics
according to geometrical parameters or through an optimisation procedure with the aim
to erase or “deactivate” low stress elements from the structure.
In the Civil Engineering field, most of the conventional reinforced concrete structures are
designed as frame systems. Ascribable to their structural configuration, the geometry of
the resistant members, and their predominant flexural performance, the global behaviour
of a structure can be accurately represented through analytic or numerical models based
on flexural beam theory. For this type of elements, the need of reinforcement can be easily
computed by determining the internal equilibrium of the resistant forces (given by the steel
and concrete) and the resulting system of local forces. On the other hand, at a local scale,
zones where the stresses due to shear are predominant over those generated by bending,
tend to develop non-flexural elements; in general, these elements are out of the range of
validity of beam theory and require a different approach to be implemented.
According to reference [Devadas, 2003], most of the cracks and failures of the structures
occur due to an inadequate attention to detailing. Often, these problems are located at
geometrical discontinuities such as joints, trims or elements presenting an abrupt change
in their thickness but also, in the zones under the effect of exceptional concentrated forces,
case of corbels and nibs. In such situations, complex stress states arise and must be taken
into account while designing the reinforcement.
As in all other zones of a structure, the main requirements are that all the existing
forces from the surroundings could be safely transmitted to the supporting members and/or
foundations. Sharp discontinuities can occur in the direction of internal forces and it is
imperative to provide a proper reinforcement able resist the tension while being consistent
with the pertinent codes in terms of the quantity, distribution, anchorage length, etc.
In many cases, standard guides found in the construction manuals like the ACI detailing
recommendations or in the EuroCodes fulfil the design but it is up to the structural engineer
to decide whether special considerations are or are not needed.
For the exceptional cases, the structural engineer should isolate specific regions, analyse
them and provide the required reinforcement computed through a convenient methodology.
In this chapter are discussed some of the most widely used methodologies for the design
of the reinforcement at non-flexural elements and, more generally, applied at disturbed
regions. The first section briefly treats the theory applied to the design of common elements
and behaviour hypothesis. The second section addresses to the use of the FEM for the
structural design. Finally, the Strut-and-Tie method is presented in the third section.
12
• The strain in the concrete is equal to the strain in the reinforcement at the same
level.
• The stresses along the element can be computed from the strains by using stress-
strain curves for each individual material (concrete and steel).
Few words must be said about the above assumptions. The first one is the traditional
“plane sections remain plane” assumption made in Euler-Bernoulli theory for beams made
of any material. The second assumption implies a perfect bonding condition between the
concrete and the steel
The third assumption needs to be attached to reference stress-strain relationships like
the ones depicted in figure 1.1. Concrete model generally consists of a parabola (equations
1.1) from zero stress to the compressive strength of the concrete. The strain that corre-
sponds to the peak compressive stress, 0 , is often assumed to be 0.002 for normal strength
concrete.
s sc
Parabola-rectangle
kfyk
A Idealised f ck B diagram
fyk
fyd
f cd
B Design
A Bi-linear
diagram
e
fyd E s eud euk e
0 e c2 e c3 e cu2 e cu3
(a) Assumed steel reinforcement stress- (b) Assumed concrete stress-strain for the
strain design of cross-sections
Figure 1.1: Assumed material stress-strain relationships according to Eurocode2.
" 2 #
c c
fc = fc0 2 − (ACI) (1.1a)
0 0
c n
σc =fcd 1− 1− for 0 ≤ c ≤ c2 [Eq. (3.17) of EC2] (1.2a)
c2
σc =fcd for c2 ≤ c ≤ cu2 [Eq. (3.18) of EC2] (1.2b)
where n is an exponent depending on the class of concrete (commonly 2), c2 is the
strain at reaching the maximum strength and cu2 is the ultimate allowed strain.
For explanatory purposes, in the last expressions fcd corresponds to the value of the
design compressive strength of the concrete defined as fcd = αcc fck /γc (Eq. (3.15) of
EC2). Being γc the partial safety factor for concrete (1.5 recommended for persistent
design situations), αcc the coefficient taking account of long term effects on the compressive
strength and of unfavourable effects resulting from the way the load is applied (taken as
1), and .
Beyond 0 , the stresses developed by the concrete are assumed to be inversely pro-
portional to strain. In tension the concrete can be assumed to present a simplified linear
stress–strain relationship up to the value of the design tensile strength, fctd , defined as:
fckt,0.05
fctd = αct [Eq. (3.16) of EC2] (1.3)
γC
For steel reinforcement, the considered model is much more simple. As depicted in
figure 1.1a, a elastic-perfectly plastic model will be assumed for the steel acting in tension
or in compression.
Even if these three assumptions allow the calculation of the behaviour of flexural RC
elements, for design purposes, additional assumptions can be made.
Figure 1.2: Simplified strain and stress distribution in a plane section [EC2].
P0 = φσc ac + as σy (1.4)
where ac represents the concrete’s cross section and φ is reduction factor that takes into
account geometric imperfections.
In real structures is not common to find elements working under pure compressive
actions. Due to accidental eccentricity in addition to the fact that almost every structure
is continuous, the axial loading and bending moments are considered together. Most
design codes recommend to take into consideration the effects of elements subjected to
flexo-compression even if the structural analysis points out zero bending forces.
Fc
a vf s
M
a q
q a
T+DT T
S S S S
Owing to the complexity of the problem, the current methods used for designing
RC elements undergoing shear are based on the results of experimental campaigns (e.g
[Mattock et al., 1976] and [Aguilar et al., 2002]). Those campaigns are principally focused
on determining the concrete resistance against diagonal cracking and the contribution of
the transverse reinforcement to the global resistance
In an attempt to predict the behaviour of elements governed by shear, Ritter [Ritter, 1899]
proposed that reinforced beam presenting inclined fissures can be idealised as a truss where
the longitudinal reinforcement acts as a tensile chord, the transverse reinforcement acts as
webs withstanding traction and, finally, the segments of concrete between the principal
cracks idealised as the webs in compression. This idealisation is depicted in figure 1.4.
The premises considered are the following:
• The fissures extend from the lowest fibber to the centroid of the compressed zone
• The self weight is neglected and distributed loads acting between cracks. In other
words, the increment in the shear, ∆V , of two sections is given by Vs where V is
the shear force in the zone between the two considered sections and s is the distance
separating them.
The analogy considers an angle θ measured between the crack pattern and the element
axis, and an angle α that corresponds to inclination of the transverse reinforcement also
compared to the element axis. In accordance to figure 1.4, the horizontal spacing between
inclined cracks and the stirrups is denoted by s. The compression force at the concrete
diagonal is Fc and the traction acting at the diagonal reinforcement is given by av fs (being
av the cross section of the transverse reinforcement and fs the force acting on it)
As result of the increment in the bending moment, ∆M , an increment in the longitu-
dinal tension, ∆T is also produced. From the equilibrium over the vertical and horizontal
forces it is found that the maximum shear force, Vmax , is given by:
av σ y z sin α
Vmax = cos α + (1.5)
s cos θ
where av represents the available shear reinforcement presenting; an array of individual
stirrups separated a distance s and tilted an angle α. σy represents the steel’s yield limit,
z lever arm (distance between the centroids of compressive and tensile chords), and T is
the force of traction acting on the longitudinal steel.
From the latter expression, it can be deduced that, if the load capacity of an element
is directly dependant on its resistance to inclined stresses, the maximum admissible load
corresponds to the development of the yielding stress at the stirrups or the transverse
reinforcement. This implies that both parts, the concrete forming the compression cords
as well the reinforcement forming the tension ones, must be able to withstand the force
increments originated by the crack evolution.
The truss analogy has been used to estimate the shear resistance of elements with
transverse reinforcement. For practical purposes and with the aim of obtaining a good cor-
relation between the calculated resistance and the experimental tests [Turmo et al., 2009]
[Rao et al., 2007], the resulting admissible load has been expressed as the addition of the
computed resistance plus the resistance of the element supposing no transverse reinforce-
ment. As expressed, this resolution achieves good correlation but lacks of theoretical
framework [Grandić et al., 2015].
Some authors have proposed modifications to the original truss model like taking into
account important factors such as the angle of cracks, transverse deformations, fan-shaped-
crack pattern, normal stress continuity along cracks to cite some of them but such modifi-
cations have not been included in the design codes [for Structural Concrete, 2008] and will
not be detailed in this document.
Given that vertical shear force has been taken to be a good indicator of diagonal
tension, diagonal tensile forces are not calculated for most of the structural elements. For
flexural elements requiring design shear reinforcement, the EuroCodes bases its hypothesis
on the truss model. Nevertheless, for beams with loads near to supports, corbels and any
element where a non-linear strain distribution exist this codes suggest to apply strut-and-tie
models.
• Numerous FEM freeware and shareware software packages conceived for structural
design and analysis are available.
• Multiple load cases can be easily programmed and added one to another according
to the principle of superposition.
For each advantage we can find one, or more, disadvantages. Some of the main draw-
backs in using the FEM are:
• Careful detailing procedures must be established and followed in order to meet the
serviceability and ductility demands.
1.2.1 Membranes
Usually, the output of linear elastic FE modelling of membranes is given in terms of stresses
and strains in the Cartesian coordinate system, whose principal axes are generally chosen
to be collinear with the envisaged preponderant reinforcement directions. The presented
solution satisfies equilibrium but presents constructive difficulties. Whilst principal stresses
are of interest as they give the elastic load path and from a simple resistance point of
view, the best reinforcement trajectory, it is usually preferable to place the reinforcement
along the axes of the structure or the structural element. For this case, the sum of the
stress resultants of concrete and steel reinforcement must equilibrate the global stresses on
the membrane. In most approaches [Collins and Mitchell, 1980], [Vecchio and Selby, 1991],
[Vecchio and Collins, 1986], the concrete is assumed to be a no-tension material and, in
the same manner, the steel reinforcement is idealised to carry no compression (unilateral
stress-strain relationship).
For a general element, the concrete and reinforcement stress resultants must sum to
the global stresses on the membrane (see figure 1.6). However, this approach does not
Figure 1.7: Mohr’s circle for reinforcing steel placed solely in the Y-direction
necessarily conducts to a unique response due to the multiple values of ρx σxy and ρy σyx
that combined can assure equilibrium .
A practical solution is to place the reinforcement along just one principal direction
[Park and Gamble, 2000]. If, for example, the designer chooses to reinforce only along the
Y direction, the prescribed solution would automatically implies that ρx = 0. Since no
stress in traction acts on the concrete (σ1c = 0), the minor stress acting on the concrete
can be obtained from σ3c = 2σ0 ; the Mohr’s circle of concrete stress state calculated from
two points on the circle (see figure 1.7) where σ0 is the normal stress at the centre of the
Mohr’s circle of concrete stresses and σx and τ xy are the X-normal and shear stresses,
respectively.
From the geometry of the circles it can also be appreciated that the required stress in
the reinforcement direction (Y -direction) is given by ρy σsy = σx + σy − 2σ0 .
In figure 1.8, the Mohr’s circle for an isotropically reinforced concrete panel is presented.
It can be appreciated that the circle representing the concrete stress state is a mere transla-
tion of the global stresses represented before (1.6) and that the quantities ρx σsx and ρy σsy
equal the major principal stress σI . It should be pointed out that the area of reinforcement
acting at each of the global axis, X and Y , is equal to the area needed for the case where
the steel is placed along the direction corresponding to the major principal stress.
The most general case of reinforcement, is given by the denominated plasticity crite-
rion [Johansen, 1962], [Nielsen, 1964] where a series of formulae (equations 1.6a to 1.6g)
2 ∂fyt ∂fyt
Y1 = τxy − (ρx − σx )(ρy − σy ) = 0 (1.6a)
∂x ∂y
2 ∂fyt ∂fyt
Y2 = τxy − (fcd − ρx + σx )(ρx − σx ) = 0 (1.6b)
∂x ∂x
2 ∂fyt ∂fyt
Y3 = τxy − (fcd − ρy + σy )(ρy − σy ) = 0 (1.6c)
∂y ∂y
2
Y4 = τxy − fcd /4 = 0 (1.6d)
2 ∂fyc ∂fyc
Y5 = τxy + (fcd + ρx + σx )(ρx + σx ) = 0 (1.6e)
∂x ∂x
2 ∂fyc ∂fyc
Y6 = τxy + (fcd + ρy + σy )(ρy + σy ) = 0 (1.6f)
∂y ∂y
2 ∂fyc ∂fyc
Y7 = τxy + (fcd + ρx + σx )(ρx + σy ) = 0 (1.6g)
∂y ∂y
where fcd is the design value of the concrete’s compression strength variations, ∂f∂xyt
and ∂f∂yyt are the design strength variations of the steel reinforcement in tension in the
global directions, and ∂f∂xyc and ∂f∂yyc are the design strengths of the steel reinforcement in
compression.
In spite of its relatively straightforward application, its applicability may not be easy
due to the difficulties related to a 3-dimensional stress space plotting and as alternative
graphic solution remains the use of Mohr’s circles of stress [for Structural Concrete, 2008].
elements design, there is much less concern with shells since such structures are typically
under-reinforced [for Structural Concrete, 2008]. In other words, the failure of this kind
of elements is governed by yielding of the reinforcement bars and not by crushing at the
concrete section. An important exception can be mentioned about concentrated trans-
verse forces, that may result in brittle punching failures without transverse reinforcement
[Unnikrishna and Devdas, 2003].
As expressed before, placing the steel bars aligned with the principal (major) stress
direction could minimise the quantity of required reinforcement for shell elements. The
so-called “trajectory reinforcement” has been applied to several slab and shell structures
in the past [?]. However, if several different load-cases must be considered, the principal
stress directions may vary from one case to another thus, making impossible to align the
reinforcement. For these and other constructibility reasons mentioned before, orthogonal
reinforcement is provided in almost all slabs and shell structures.
The output of a shell element consists in eight independent stress resultants: bending
and twisting moments (Mxx , Myy Mxy , and Myx ), transverse shear forces (Txz and Tyz )
and membrane forces (Nxx , Nyy , and Nxy ) 1.10a. By applying equilibrium equations over
the body, it can be stated that:
∂Txz ∂Tyz
+ +q =0 (1.7a)
∂x ∂y
∂Mxx ∂Mxy
+ − Txz = 0 (1.7b)
∂x ∂y
∂Myy ∂Myx
+ − Tyz = 0 (1.7c)
∂y ∂x
Moment equilibrium equations yields to:
at steel reinforcement and maximum deflection occurring at the centre of the element. At
this stage, negligible cracking may occur at the lower layer, zone where the concrete’s
tensile capacity will be exceed due to the forces carried out by flexural behaviour (figure
A.6c). Increasing the load beyond the service limit, will increase the size and depth of the
cracks and may induce the yielding of the reinforcement. Increasing the load still further,
will propagate the cracks to the free edges of the element generating the yielding lines to
cross all reinforcing bars. At this ultimate state, the yield lines form boundaries and allow
rotation between the rigid or "intact" parts, thus creating mechanisms and the instability
of the element (figure 1.9b).
The ultimate load of concrete slabs and shells has been investigated by considering
local stresses and strains within the element and their corresponding yield conditions under
the basis of plasticity theory and flow rules for concrete and for the inner reinforcement
[Nilson, 1997]. This approach gives accurate results for almost any case but its application
is rarely justified.
By superimposing the ultimate shell’s moments Mxu and Myu along the global rein-
forcement directions and setting Mxy , Nxx , and Nyy equal to zero, a simplified statically
admissible state of stress is obtained. Now, for an arbitrary direction n, and based on
equations 1.8, it can be stated that Nn = Ntn = 0, Mn = Mxu cos2 φ + Myu sin2 φ and
Mtn = (Myu − Mxu ) sin φ cos φ. For most cases, the resultant depths of the compression
zone in the concrete section do not coincide for the two orthogonal directions, cx 6= cy . As
a result, there is no development of compatible mechanism (figure 1.10c). The difference
between the value obtained for Mn and the value of Mnu for cx 6= cy have been found of a
negligible order leading to:
0 0 0
Mnu = Mxu cos2 φ + Myu sin2 φ (1.11)
From equations 1.8, a state of stress in terms of Mxx , Myy , and Mxy corresponds to
bending and twisting moments in direction n is given by Mn = Mxx cos2 φ + Myy sin2 φ +
Mxy sin(2φ). From combining these previous expressions with the inequality condition,
−Myu ≤ Mn ≤ Myu , the yield conditions for orthogonally reinforced shell elements can be
Figure 1.10: Normal yield criterion: a)shell element, b) Yield line, c) superimposition of ultimate moment in
directions X and Y d) Yield condition e) dimensioning.
2
Y = Mxy − (Mxu − Mxx )(Myu − Myy ) = 0 (1.12a)
Y 0 = Mxy
2 0
− (Mxu 0
− Mxx )(Myu − Myy ) = 0 (1.12b)
0
where some restrictions are applied; (Mxu − Mxx ≥ 0, Myu − Myy ≥ 0, Mxu + Mxx ≥ 0,
0 0
and Myu + Myy ≥ 0. In figure 1.10, the conditions Y0 and Y = 0 are represented.
Finally, the yielding reinforcement layers are generally substituted by an equivalent
orthogonal reinforcement. Hence, fictitious resistances, Nxs , Nys and Nxys , are computed
in order to take into account the effect of the reinforcement distributed into several layers
oriented in directions differing by angles Θi measured from the X-axis and with individual
resistances equivalent to its cross section, Nis = (as fsy )i per unit width.
Nis cos2 Θi
X
Nxs = (1.13a)
i
Nis sin2 Θi
X
Nys = (1.13b)
i
X
Nxys = Nis cos Θi sin Θi (1.13c)
i
V0
Mxy Nxy
-
dv dv 2 dv dv
M
- xx + Nxx
Nyx Nyy dv 2 1
Nxx Nxy Tyz
Myy V0 cotq V0 cotq
Tyz Txz q 2
Txz V0
V0 cotq
Mxx Myy Nyy dv cotq
+
dv 2
Mxx Nxx Mxy Nxy
+ dv
+ V0 cotq
2 2
dv 2
Due to its simplicity, the normal yield criterion is widely used for the design of concrete
slabs in current practice [Kennedy and Goodchild, 2004]. Nevertheless, this method is not
conceived for elements presenting excessive reinforcement ratios [Marti, 1978].
Mxx Nxx
Nx inf,sup =± + (1.14a)
dv 2
Myy Nyy
Ny inf,sup = ± + (1.14b)
dv 2
Mxy Nxy
Nxy inf,sup = ± + (1.14c)
dv 2
On the contrary, if the shear stress exceeds the nominal concrete’s shear cracking resis-
tance, the core fissures and is treated as the web of a girder of flanged cross-section along
direction φ0 (figure 1.11). The inclination produces tensile forces that must be resisted by
the covers modifying equations 1.14 as follows:
where a k and k 0 are arbitrary positive factors (normally taken equal to 1), θ is the
inclination of the diagonal compression, as and a0s are the bottom and top reinforcement
areas per unit width.
For the case where the core is cracked, the need of transverse reinforcement ratio is
computed by ρ = V0dvtan
σy
θ
. For the opposite case, the terms containing Txz or Tyz can be
ignored from equations 1.16 and it is assumed that no transverse shear reinforcement is
needed.
Aiming to prevent yielding in all directions, this method considers the Johansen’s yield
criterion (normal moment yield criterion); at any point in the slab, the moment normal to
any given direction, n, due to design moments Mx , My , and Mxy (figure 1.12), must not
exceed the ultimate normal resisting moment in that same direction. The ultimate normal
resisting moment is typically provided by ultimate resisting moments Mux and MuΘ related
to the reinforcement in the X and Θ directions.
M yx
Myy
Y
Mxx Mxy
Based on the principles of the plate-type behaviour and the consideration of solid con-
crete elements reinforced with unidirectional layers of reinforcement oriented along the
global axis, the W&A method considers a regular geometry element subjected to a mo-
ment field (Mx , My , Mxy ). The reinforcement is considered to undergo only tensile forces
developing a resistant stress σs while the compressive forces are taken by the concrete.
The procedure attempts to find a feasible solution to reinforce along the principal axis.
For this purpose, a transformation of the moments over the principal axes is needed and
then, the requirement is ensured by calculating the resisting moment M ∗ and the actuating
moment as a function of the triad of acting moments according to equations 1.17 to 1.22
for the top and the bottom reinforcement [Wood, 1961]. Two types of design moments M ∗
are calculated then calculated. The lower (positive) and the upper (negative) moments
(respectively causing mainly tension in the bottom parts and in the upper parts).
Based on these concepts, the reinforcement at the bottom of the slab in both directions
must be designed to provide positive bending moment resistance in an X-Y system and
compared to the transformed moments acting in an n−tsystem. When the external normal
is at an angle Θ measured clockwise from the X-axis, the general transform is as follows:
For the lower reinforcement
(M + Myy cot Θ)
xy
Mx∗ = Mxx + 2Mxy cot Θ + Myy cot 2
Θ + (1.17)
sin Θ
Myy (M + M cot Θ)
xy yy
MΘ∗ = 2 +
(1.18)
sin Θ sin Θ
∗
If Mxx <0: (1.19a)
Mx∗ = 0 (1.19b)
2
(Mxy +Myy cot Θ)
Myy + Mxx +2M
2
xy cot Θ+Myy cot Θ
MΘ∗ = (1.19c)
sin2 Θ
As expected, while considering previous equations for the upper reinforcement, the sign
of of the last term must be inverted [Clarke and Cope, 1984].
In a similar manner, the effects of the in-plane forces (membrane effect) are considered
as follows:
(N + Ny cot Θ)
xy
Nx∗ = Nxx + 2Nxy cot Θ + Nyy cot 2
Θ+ (1.21)
sin Θ
Nyy (N + N cot Θ)
xy yy
NΘ∗ = 2 +
(1.22)
sin Θ sin Θ
∗
If Nxx <0: (1.23a)
Nx∗ = 0 (1.23b)
2
(Nxy +Nyy cot Θ)
Nyy + Nxx +2N
2
xy cot Θ+Nyy cot Θ
NΘ∗ = (1.23c)
sin2 Θ
T yz
M yx
Nyy Myy
T xz
Y
Nyx
Nxy
Nxx
Mxy
X
Mxx
This method acts as a post-processing step conducted over the state of generalised
stresses previously obtained on a FE calculation. The problem is to determine the optimal
economic values of the longitudinal reinforcement sections, as in the upper layer, U , and
the lower one, L. In order to attain that result, the algorithm is based on the principle of
the equilibrium of different facets centred at the point of calculation whose normal rotates
in the plane tangent to the average sheet.
For each one of these facets, the bending moment ({M }) and the membrane tension
({N }) are applied according to the current stress state tensors and are evaluated using the
following equations:
Taking into account the resultant bending moment and the membrane forces, the mo-
ment of service which steel must take can be evaluated. By a combined compression-and-
bending calculation, it is now possible to determine the lower tensile forces Φ(θ) and higher
Φ0 (θ), perpendicular to the section, which must be balanced by the bottom and top layers
of reinforcement.
The resisting forces, Φ0∗ , in the direction θ of the two plies can be evaluated using the
following expressions:
Φ∗ (Θ) = aXL cos2 θ + aY L sin2 θ σy (1.28)
for the lower layer
Φ0∗ (Θ) = aXU cos2 θ + aY U sin2 θ σy (1.29)
for the upper layer
where σy represents the maximal admissible stress in the steel (identical for both direc-
tions).
Considering that the resisting forces Φ∗ (Θ) and Φ0∗ (Θ) must be greater than the applied
ones, Φ(Θ) and Φ0 (Θ) respectively, the optimum of the reinforcement corresponds to the
minimum quantities of:
Φ0 (Θi )
subject to: aXL cos Θ + aY L sin Θ ≥ (1.31a)
σ
aXL ≥ 0 (1.31b)
aY L ≥ 0 (1.31c)
(1.31d)
a ys
F' (Q i )
s sin2 Q i Validity
domain
a xs
0
F '(Q i )
s cos2 Q i
Figure 1.15: Validity domain (adapted from [Capra and Maury, 1978]).
a ys
2
2
as
a*ys A
C
a xs
0 a*xs
Figure 1.16: General validity domain for different values of Θ (adapted from [Capra and Maury, 1978]).
|M | N
y−
σc,max = (1.32)
I A
For transverse reinforcement, the proposed calculation starts from the equivalent shear
stress expressed as:
1q 2 2
τ=
Txz + Tyz (1.33)
2
where z represent the arm of lever of the elastic couple of the section and TZX and
TZY are the stresses cutting-edges. Thus, the section of transverse reinforcement is simply
obtained by dividing this constraint by the acceptable ultimate stress of steel.
1.2.2.6 Remarks
Laboratory test campaigns carried out on slabs elements have shown that the use of yield
line theory leads to conservative designs [Marti et al., 1987]. The same tests have been
numerically reproduced confirming previous results [May and Ganaba, 1988]. Normally,
yield line analysis supposes mechanisms that ignores possible membrane effects. In other
words, most mechanisms are not kinematically admissible for a section in which the neutral
axis is not at the centre of the section. In addition, resent literature [May and Lodi, 2005]
points out that conservatism carried out by the method decreases or even disappears as
the area of reinforcement increases.
On the other hand, transverse shear forces obtained from a FE model will generally
be less accurate than bending and twisting moments, since, transverse shear forces are
calculated as derivatives of the bending and twisting moments (see equations 1.8). Thus,
a relatively fine mesh would be required in zones potentially presenting important shear
forces; however, in practice, mesh sizes and geometries are commonly dictated by the size
of particularly large overall building models.
In the current industry of nuclear civil works, meshes presenting elements from 0.5 to 1
meter side are typically used to estimate reinforcement ratios [Herve et al., 2014]. Further-
more, the intensive use of surface elements may lead to poor representations of geometric
discontinuities; e.g. abrupt thickness transitions, heterogeneous mid-plane positions and
joints.
One of the ways to evaluate the quality of the mesh is to compare results to test data
or to theoretical values. Unfortunately, test data and theoretical results are often not
available. So, other means of evaluating mesh quality are needed. These include mesh
refinement and interpretations of results discontinuities.
The most fundamental and accurate method for evaluating mesh quality is to refine
the mesh until a critical result, such as the maximum stress in a specific location converges
(i.e. it doesn’t change significantly with each refinement). Another option is to evaluate
the magnitude of stress discontinuity between adjacent elements in the critical region. In
most cases, the finite element method computes stresses directly at interior locations of
the element (Gauss points) and extrapolates them to the nodes on the element boundaries.
While it is common to view these stresses as average values, the reality is that each element
calculates different stresses at shared nodes.
Figure 1.17a exposes an overview of the thicknesses of the different structural parts.
Figure 1.17b indicates the relative thicknesses of the different shells that are used.
Figure 1.17b delivers a typical modelling using only shell elements where the position of
the mid-plane of the shell elements may present huge differences with the real structure.
Figures 1.18a and 1.18b, show a thermal load case the drawback only for the thermal
gradient description even before considering the associated generalised forces or stresses
that such a modelling generate compared to the brick modelling.
analysis can also be used for evaluating complex geometries or poorly detailed structures
presenting effects of localised cracking.
One common application of non-linear analysis is the confirmation of safety for complex
design details.
In the industrial practice, and due to the associated cost, this type of analyses are
implemented in situations specific situations:
Even if the principle can be directly applied to elements resisting flexion with a linear
distribution of stresses [Hsu, 1992], this document will focus only on its application for
detailing discontinuity regions undergoing no-linear distribution of the stresses through
the member depth.
1.3.2 Fundamentals
A ST model design adheres to two principles: 1) the resultant truss model must be in
equilibrium with the external force system and, 2) the concrete element has enough defor-
mation capacity to accommodate the assumed distribution of forces [Schlaich et al., 1987].
Proper anchorage length of the reinforcement is an implicit requirement in order to assure
the needed ductility. Complementary, the compressive stresses developed in the concrete
must not exceed the factored concrete strengths, and the tensile stresses the factored steel
capacities. If all of the mentioned above requirements are satisfied, the application of the
STM should result in a conservative design [Williams et al., 2012].
As mentioned before, the STM consists of three principal components: struts, ties, and
nodes. The compression members, referred as struts, are considered to be made out of
concrete Cc ; the tension members, referred to as ties, may be made of concrete without
reinforcement Tc (case not considered within this work) or reinforced by layers of mild steel
reinforcement or prestessing steel Ts . The struts and the ties intersect at regions referred to
as nodes. Due to the concentration of stresses from intersecting truss members, the nodes
are the most highly stressed regions of a structural member and their revision should be
considered during the overall analysis.
Global model
Region discrimination
Element design
Realistic
no
steel dis-
tribution
yes
Keep
no model
yes
Arrange reinforcement
Figure 1.21: Flowchart design procedure using STM (inspired by [El-Metwally and Chen, 2017])
F3 F4 F5
F2 F6
h1 D region h2 Ma
Mb F1
h2 h2 F7
F9
F8
(a) Global element (b) D-region and boundaries (c) Isolated model
Figure 1.22: Local zone of a beam with sudden thickness change
• Punctual loads acting on the structure at a very close proximity to each other may
be resolved together to simplify the development of the strut-and-tie model. The
decision whether or not to merge loads together is left to the designer.
• A distributed load must be divided into a set of punctual loads acting at the nodes
of the truss system. The self-weight of the structure must be applied to the STM in
the same manner.
In some occasions, each load case will create a unique set of forces causing the locations
of the critical regions of the STM to change. Therefore, depending on the load cases, a
ST analysis should be performed for each individual load case generating as many truss
systems as load cases (see figure 1.23).
Geometry proposal. Based on the equivalent punctual loads derived from the previous
steep (initial nodes), the designer projects the struts and the ties as straight elements
from one node to another with the aim of developing a stable structure within the solid
element. This process may or may not include the interaction of all the initial nodes
and, in addition, it may include secondary nodes depending on the designer’s criteria. For
this step, a linear finite element model is commonly used in order to visualise the flow of
forces within the member; and align the truss elements according to the stress trajectories
[Schlaich et al., 1987].
Analyse ST models and member design. At this point, the ST model corresponds
to a planar (or spatial) truss depicted by a nodal list, an element list and, an inter-
connectivity table. Using these three items, a mathematical model can be then built and
solved. The forces developed in the elements are determined by solving a model including
the previously determined boundary conditions and factored system of loads. The area
of material needed for each element in the STM should be sufficient to safely resist the
computed force without surpassing the yield strength of the steel nor the limit resistance
in the concrete. In a conventionally reinforced structure, the area of reinforcement needed
for a tie, Ast , is determined from the following equation:
Fu
Ast = (1.37)
φR
where:
Fu corresponds to the largest force in the element for all load combinations considered,
R the material’s resistance (σy for steel, σc the resistance for concrete material) and,
φ the safety factor (0.9 for traction [Eurocode2, 2008, ACI-318, 2008])
Additionally, the Eurocodes requires the diminution of the design strength for struts in
cracked compression zones:
Figure 1.24: Nodal proportioning techniques - hydrostatic versus non-hydrostatic nodes [Birrcher et al., 2009]
referred as CCC, CCT, CTT or TTT (C standing for compression forces and T standing
for tension ones). Due to geometric complications, it is highly recommended to merge
groups of forces in order to reduce their number If more than three elements intersect at
one node. However, this is not always possible and often generates other possible type of
nodal combinations as showed in figure 1.25.
For hydrostatic nodes, the faces are perpendicular to the attached elements and the
length sides are proportional to the strut forces. In the presence of ties, position of the
faces are proposed by assuming the tie forces act from behind the node to compress the
nodal region.
Table 1.1: ACI and Eurocode values for nodal zones resistance
σRd,max
Nodal zones, σRd,max
ACI Eurocodes
0
CCC type nodes 0.85(1)βs f c (A.5.2.1) 1ν 0 fEcd (6.5.4)
Nodal zones anchoring a tie 0.85(0.8)βs f 0 c (A.5.2.2) 0.85ν 0 fEcd (6.5.4)
In nodal zones anchoring two or more ties. 0.85(0.6)βs f 0 c (A.5.2.3) 0.75ν 0 fEcd (6.5.4)
Non-hydrostatic nodes are proportioned based on the origin of the applied stress. In the
case of CCC non-hydrostatic nodes, the approach is to set the back face dimension as the
effective depth of the compression block. In the presence of a CCT, the back face dimension
is taken as twice the distance from the centroid of the longitudinal reinforcement to the
extreme tension fibre of the beam. Regarding the nodes located at the boundary conditions
and supports, the dimension of the bearing face is determined by the dimensions of the
bearing plate. This proportioning technique allows the geometry of the nodes to closely
correspond to the actual stress concentrations at the nodal regions. In contrast, the use
of hydrostatic nodes can sometimes result in unrealistic nodal geometries and impractical
reinforcement layouts [Birrcher et al., 2009].
Figure 1.27: Parameters for the determination of transverse tensile forces in a compression field with smeared
reinforcement [Eurocode2, 2008].
• The distribution of the truss elements can be based on the stress distribution of a
prior elastic FEM analysis [Schlaich et al., 1987], [Eurocode2, 2008].
• Align the struts with ±15o of the stress trajectories to represent the nature of the
element [Ramirez and Breen, 1991].
• The most efficient ST models are typically those with the fewest and shortest ties
[Williams et al., 2012].
• The ties should coincide in position and direction with the corresponding reinforce-
ment [Eurocode2, 2008].
1.3.5 Remarks
The most important benefit of ST modelling is its versatility. Almost any structure can
be treated with this method in order to obtain a conservative design. In contrast, the
flexibility of its application is granted by the lack of explicit guidance and consistency in
the current codes [Barton et al., 1991].
1.5 Summary
In the field of engineering practice, the construction codes allow the analysis and design
of non-flexural RC elements trough 3 principal approaches: recommendations based on
empirical relations, FEM based algorithms and also rational methods.
Even if several campaigns have been carried out with the aim of determine the behaviour
and the resistance of elements such nibs and deep beams, the results remain valid for a
small range of cases where the geometry of analysed element as well of the load system
and steel disposition resembles to those studied restricting its application for particular
cases. When a special case is confronted, a FE analysis is, probably, the first alternative
to manual calculation.
As expressed before most structural designs are performed using FEM software and
post-processed trough automatic algorithms. Most of the applied methodologies were con-
ceived to treat elements where the effects carried out by a flexural behaviour predominate
over those produced by shear, limiting their range of validity if no special considerations
are implemented. It is also worth to mention the fact that, in the most popular methods,
W&A nor C&M, no explicit check of the assumed failure mode is made at the elements de-
signed. These methodologies implicitly assume that the reinforcement ratios obtained are
small enough to avoid concrete crushing. While this condition is usually satisfied in slabs,
a check of the assumed failure mode should always be carried out for shells, particularly
for elements subjected to significant axial compression or cases with high reinforcement
ratios.
Despite the fact of the existence of some other FE-based methods such as the sandwich
or a full three-dimensional solid design do indeed take into account a verification on the
concrete, their application has not been included within the construction codes until this
day.
Other alternative is the use of the rational methods that can be seen as an intermediate
point between structural mechanics and the empirical methods. The main drawbacks of
the use of such methods has been their lack of explicit guidance and consistency in the
current codes, aspects that propitiate a strong dependency on the designer’s expertise.
The guidance provided by the elastic stress field, result of a prior FE analysis, ensures
a good behaviour at the serviceability limit state and can also be used to study its ultimate
limit state response. The procedure seems to be straightforward for cases where the ties
are aligned to the orthogonal reinforcement; for cases where this does not occur, the final
reinforcement may not present the desired behaviour
The performance of a model developed through rational methods is totally dependent
on the intuition and practice of the structural engineer as well as the expertise on the
problem to solve. The modifications and decisions taken within the design process becomes
a repetitive and logic task for common geometries and load cases but, it can also be a rather
difficult task when facing unusual cases.
The modification of a model in order to adapt it to different requirements, or just in
order to improve it response, requires a knowledge on the relation of the mechanical char-
acteristics and the evolution on the structural behaviour. In a conventional process, any
modification is dictated by preexisting codes and construction recommendations. For STM
these rules can sometimes depend on the typology of the problem and even be subjective
rules presenting guidance for a series of solved recurrent examples but needing a sort of
extrapolations for different problems.
In recent years, the development of Computational Aided Design (CAD) tools allows
to include rational techniques intended guide the engineer through the design process. Dif-
ferent techniques adapting mathematical methods have been proposed in order to achieve
optimal solutions for automatically computing the reinforcement of concrete structures
though ST models. Next chapter intents to make an overall revision of the most common
optimisation techniques and algorithms that have been applied to propose and to justify
the trajectory of ST models.
The selection of the "best" Strut-and-Tie model has always been a concern. In addition,
among the actual market demands are those aiming for lower costs, shorter time and higher
quality standards that, have forced the designers to choose methodologies that allow the
reduction of costs in the stages of conception and implementation.
Within the application of the STM, one of the principal difficulties have been to over-
come the manual tasks and decisions based on thumb rules.
Some researchers, such [Alshegeir and Ramirez, 1992, Kuchma and Tjhin, 2001], are
focused on developing tools and criteria to aid engineers to perform the manual tasks and
to provide information to take decisions. Nevertheless these approaches allow the designer
to provide satisfactory models with relative ease, the manual trial-and-error-method used to
compare one model to another is still present. In contrast, other groups of research incline
their studies to fully automatise the process based principally in structural optimisation
methods.
In order to fulfil an appropriate Strut-and-Tie model, the design of a structural concrete
member can be transformed into a structural design optimisation problem. The solution
of such problem can be then solved by different methods or techniques depending on the
desired objective and the imposed constraints.
This second chapter presents a review of academic literature, in the areas where the
present work is contributing, namely structural optimisation and computer aided tools for
Strut and Tie models.
The first part concerns the methodologies applied to the structural optimisation while
the second and third parts deal with the practice of the structural optimisation to compute
ST models.
47
• inequality constraints
• equality constraints
• side constraints
Inequality constraints are those which impose either a lower or an upper limit on some
quantities. Equality constraints represent specific requirements of a quantity taking a
selected value while, the third type, the side constraints, impose at the same time an
upper and a lower restriction.
with
x1
x2
X=
..
.
xn
where X denotes the vector containing the design variables. Commonly, minimisation
problems are preferred to maximisation ones. However, a maximisation problem can be
solved indirectly by minimising the problem with opposite value. In a similar way, the
less-than constraints can be transformed into greater-than ones by multiplying them by
−1 affecting only the sign convention in some of the final results.
When using constraints that do not correspond to the same order of magnitude (e.g.
displacements and lengths) it results advantageous to represent them in a normalised form:
xm
i − xi
xi ≤ xm
i becomes ≤1 (2.3)
xm
i
• Sizing optimisation: the design variables are the thicknesses or the cross sections of
the elements of a predefined structure. Generally, the design variables are governed
by an inferior constraint of non-zero values or replaced by a greater-than relation
(see figure 2.1a).
• Shape optimisation: in this case the structure is intended to be improved through
modifications in the form, the contour of some part of the boundary of the structural
domain. Shape optimisation does not alter the connectivity of the structure: new
boundaries are not formed (see figure 2.1b).
Figure 2.1: Three categories of structural optimisation: a) Sizing optimisation of a truss structure, b) shape
optimisation and c) topology optimisation [Bendsøe and Sigmund, 2003].
• Topology optimisation: This is probably the most general form of structural opti-
misation. As in the sizing optimisation, the design variables are the value of the
thicknesses or cross sections but this time allowing them to reach zero values. If pure
topological features are optimised, the optimal values of the design variables should
take only two values: 1 (the element exists) and 0 (the element is absent) (see figure
2.1c).
A fourth type of optimisation arises when the selection considers some characteristics
of the material such as E, σy or ν. When treated as a continuous variable, the material
selection can be used to study non-conventional materials such as polymers or fibre-matrix
mixtures. On the other hand, due to the nature of conventional materials, this kind of
problem may lead to optimal solutions that points out a non real material. When treated as
discrete variables, the complexity of the problem may considrrable be increased increased
[Kirsch, 2012]. Given the vast quantity of possible materials and in order to reduce the
complexity of the problem, a small list of possible materials must be specified.
X2 X2
X2
Local minimum
Local minimum Local minimum
Feasible
Region
X1 X1
X1
Global minimum
Minimum
X1 Minimum X1 X1
A minimisation problem is referred as convex only if the objective function is convex and
the constraints bound a convex domain.
Svanberg [Svanberg, 1981] identifies three principal constraints that lead to convex
problems in structural optimisation:
where uUi is an upper limit for the displacement at the ith node.
max ui ≤ uU (2.5)
• Lower limit of the smallest eigenvalue. The constraint places a lower bound, λm , on
the smallest of the N eigenvalues of the structural stiffness matrix K
λm ≤ λj j = 1, ..., N (2.6)
Common constraints such stress or displacement related ones produce non-convex re-
gions thus, the feasible regions for realistic structural optimisation problems are expected
to be non-convex which is not always easy to verify. In cases presenting two or three
variables, the design space is reduced to a plane or to a three-dimensional space. In the
general case of n variables the design space becomes a n-dimensional hyperspace
! ! !
∂F ∂h ∂F ∂h ∂F ∂h
+λ dx1 + +λ dx2 + ... + +λ dxn = 0 (2.10)
∂x1 ∂x1 ∂x2 ∂x2 ∂xn ∂xn
In order to vanish the elements inside each parenthesis one must determine λ which
leads to a system of n equations and n + 1 unknowns (the n design variables plus the
Lagrange multiplier). The additional equation needed to solve the system is given by the
constraint relation (h(x) = 0). If multiple constraints must be treated, additional Lagrange
multipliers shall be added for each of the constraint functions. The general formulation
of an optimisation problem of one objective function, n design variables and, ne equality
constraints is thus equivalent to a n unconstrained problem with an auxiliary function
(equation 2.11). Then the optimum design can be found by solving the system formed by
equations 2.12.
ne
X
L(X, λ) = F (X) + λj hj (2.11)
j
∂L
= 0, i = 1, ..., n (2.12a)
∂xi
∂L
= 0, j = 1, ..., ne (2.12b)
∂λj
[K]{U } = {F } (2.13)
where [K] is the global stiffness matrix, {U } is the displacements and, {F } the applied
load vector.
The total weight, W , of a structure composed of m elements, can be calculated as:
m
X
W (ai ) = ρi ai li (2.14)
i=1
where ρ is the mass density and ai li the volume of the ith element. The element length,
li , is considered constant while the cross sections, ai , are taken as the design variables. The
generalised constraints gj (X) applying to the structure can be written as:
with Eij = {r}i [k]i {sj }ai (or Eij = Fi Uij li /Ei for truss structures). In the previous
expressions {r}i and {sj } are the displacement vectors generated at the ith element by the
a vector {R} and a virtual load vector {S j } associated to the jth constraint, Fi is the force
in the ith bar.
Given the objective function to minimise, W (ai ), and the constraints, gj (ai ), the La-
grange multipliers can be applied to the minimisation in the form:
m
X p
X
W (ai , λj ) = ρi li ai + λj gj (ai ) (2.17)
i=1 j=1
where λj are the Lagrangian parameters. Now, the local constrained optimum is ob-
tained by differentiating the previous equation (2.17) with respect to the design variables
ai resulting in equation 2.18.
p
∂ X ∂
W (ai , λj ) = ρi li + λj gj (ai ) = 0 i = 1, ..., m (2.18)
∂ai j=1 ∂ai
with λj ≥ 0 and λj gj = 0.
Substituting equation 2.16 into 2.18 leads to:
p
X Eij
λj =1 i = 1, ..., m (2.19)
j=1 pi li A2i
The optimal structure must satisfy the optimality condition (equation 2.19) and the
constraints (equation 2.15). When solving problems whith only one constraint, the La-
grange multiplier can be explicitly defined and used to derive the recurrence relation for
the design variable. When there is more than one constraint, the problem becomes more
complex.
Since the obtained equations are non-linear, the solution schemes are based on the use
of recurrence relations iterative in nature. In few words, the optimality criterion is used
to derive a relation to modify the design variables while the constraints equations are used
to obtain relations for the evaluation of the Lagrange parameters [Khot and B., 1979]. For
truss structures, the allowable stress constraint can be directly replaced by a constraint on
the deformation of each element.
λj ≥ 0 j = 1, ..., J (2.21)
A 2D graphical representation of the KKT conditions is given in figure 2.3. In the first
situation, where −∇F is not within the zone delimited by ∇gi , the value of F can still
be improved without violating the constraints. Thus the point is not a minimum. In the
second situation, the vector −∇F lies in the zone defined by the constraints.
Based on equation 2.20, and knowing that the first term of the left-hand side cannot
∂F
be zero ( ∂x i
6= 0), it can be stated that an equivalent expression for finding the minimum
based on an iterative process is obtained by:
m
,
X ∂gjk ∂Fk
− λjk = Tik (2.22)
j=1 ∂xi ∂xi
where Tik will take a value of 1 at the minimum
Equation 2.22 is used as an indicator of proximity to the point of minimum and, ac-
cording to the value of Tik , a recurrence relation can be established of any element xi of
design variables k X and k+1 X at two consecutive iterations. A common recurrence rela-
tion [Khot and B., 1979] and expression to compute the Lagrange multipliers are shown in
equations 2.23 and 2.24.
k+1
xi = [α + (1 − α)(k Ti )]k xi i = 1, ..., n (2.23)
n m n
∂(k gj ) k X ∂(k gj ) ∂(k F ) k
∂(k gj ) k
,
X gj X
( xi ) λj = − ( xi ) (2.24)
i=1 ∂xi j=1 ∂xi ∂xi 1 − α i=1 ∂xi
where α is the relaxation factor that limits the step size between iterations. The usual
value of α oscillates between 0.5 to 0.75 [Adeli, 2002].
Equation 2.24 represents a system of linear equations with non-negative solutions for
all λi associated to the active constraints.
As it can be inferred, finding the values of the Lagrange multipliers is trivial to solve
the optimisation problem and this task that can be achieved through an iterative process.
A general process can be summarised in four steps:
2. Computation of the Lagrange multipliers λi , task that can be achieved through equa-
tion 2.24
minimise F = CT X (2.25)
where the vectors b1 , b2 , and C and the matrices A1 and A2 are all parameters whose
dimensions match the dimensions and X, hk and gj . For these cases, the objective function
contours are straight intersecting lines and the solution will be found at the intersections
of the active constraints.
Probably the most popular method to solve linear problems is the well-known Simplex
Method (including its revisions) [Dantzig, 2016]. This method, tests adjacent vertices of the
feasible region in a sort of sequence so that at each new vertex tested, the objective function
improves or unchanges but does not decrease. The simplex method has been presented as
an efficient tool [Wright and Nocedal, 1999], generally taking 2m to 3m iterations (being m
the number of constraints) to converge. However, the effort to obtain the solution increases
exponentially for the worst case [Klee and Minty, 1970].
Briefly explained, the simplex algorithm minimises a linear objective function (equation
2.25) subject to equality constraints (equations 2.26)
The first step consists in transforming the problem into a system of linear equations
of a standard form. The objective function takes the form of f (X) = 0 and slack vari-
ables, whose value is constraint to be equal or greater than zero, are added to the current
constraint equations.
The solution is achieved by starting from an initial "guess" and improving through an
iterative process based on the Gauss-Jordan procedure.
Other type of procedures for LP are the interior point methods. These type of methods
construct a sequence of strictly feasible points lying in the interior but not belonging to the
boundaries in order to converge to the solution. Interior point methods such as Karmakar’s
or Mehrotra’s ones [Karmarkar, 1984, Mehrotra, 1992] have been proved effective but the
experience with their application to real problems is still very limited [Hernández, 1993].
k+1
X =k X + α(k S) (2.27)
x Contours of the
objective function
f (x)
Third
constraint
Improving feasible
directions
Second
constraint
Fourth
constraint
Considering a NLP problem, the Zoutendijk algorithm consists in the following steps:
minimise z (2.28)
∗ ∗ ∗
The process continues according to the solution vector k S and value k Z. If k Z = 0,
stop. Otherwise the algorithm moves to step 2
2. Step k α. The step is then proposed to be the solution of the following line search
problem:
The main drawback that the MFD presents is that it does not present a closed al-
gorithmic map [Bazaraa et al., 2013]. This can sometimes lead to solutions that do no
lie in the feasible region, violating one or more constraints, and hence presenting prob-
lems of convergence. Some modifications have been proposed to overcome this problem
([Topkis and Veinott, 1967, Vanderplaats, 1984a, Vanderplaats and Moses, 1973]), one of
the simplest consisting on the addition of the logic condition:
2.1.7 Convergence
The results obtained by the implementation of numerical algorithms are sequence of points
vectors idealised as a generalised function or mapping. In short, the application of an
algorithm A over an arbitrary set of points k X = {k x1 ...k xm }T would generate a sequence
k+1
X:
k+1
X = A(k X) (2.32)
To sum up, global convergence is directly related to reliability (or robustness), while
local convergence corresponds to efficiency on the applied algorithm.
2.1.7.1 Termination
A common optimisation algorithm tends to reach the solution when the error, , equals zero
but this solution is not always reachable nor necessary. The iterative process is terminated
when no improvement can be done on the value of the objective function without violating
the constraints. However this is not the only possible solution. Some methods stop the
iterative procedure when the progress on the improvement becomes slow, while others
base their decision on KKT conditions [Arora, 2004]. The criteria of stopping the iterative
process should correspond to the following five situations:
Ideally the best way to determine to stop or not the process would be a test comparing
directly the current values and the optimal ones |k xi −∗ xi | ≤ i , where i are small values
determined by the user. Given that for most cases these values corresponding the optimal
solution are not known, a useful test is:
||∇k f || ≤ (2.35)
Precedent equation (2.35), together with a satisfactory positive-definiteness of the Hes-
sian evaluated at k X, seems to be a reliable index for the termination with the disadvantage
of the restiveness of its application to problems with large numbers of variables.
When solving constrained problems, apart from convergence tests, an acceptable con-
straint violation is allowed by setting:
||gj || ≤ (2.36)
being gj the vector off all active constraints.
In reality, determining the proper termination criteria seems to be a matter of experience
and expertise and is itself a subject of study [Kirsch, 2012].
continuous variables along the admissible domain. The optimisation consists in the
deciding whether each element should contain material or not. To achieve this, the
material within each element is associated to a design variable defined between 0
and 1; 0 considers a void or a very weak material while 1 represents a solid material
[Bendsøe and Sigmund, 2003].
The final result can be graphically represented showing a rough description of the
optimum outer and inner boundaries (see figure 2.5).
2. Geometrical continnum optimisation or Macroapproach. This class of optimisation
proceeds with a FE analysis considering fixed characteristics for the materials of the
elements associated to a non-fixed mesh. The topology of the optimum structure is
achieved by growing/reducing material or inserting holes. The first method starts
from the hypothesis that the optimal design is obtained as a subset of the design
domain. The second method iteratively proposes holes at specific points on the
structure continuously modifying the boundaries of the geometry.
Even if some articles have been published showing Strut-and-Tie models found trough
continuum-type optimal topology, [Almeida et al., 2013, Shobeiri, 2016] this practice re-
mains controversial. From the point of view of mechanics, the characteristics of the dis-
crete and continuum structures are very different, and from the point of view of the prac-
tical implementation, there is no criteria for constructing truss topologies from the results
of optimal finite element solutions [Starčev-Ćurčin et al., 2013]. In addition, due to the
characteristics presented by the interaction concrete-reinforcement and according to some
studies [Swan et al., 1999], a more adequate ST model would be found through a discrete
optimisation purely based on truss structures.
The methods based on mathematical programming can also be found in the literature
[Lamberti and Pappalettere, 2000]. A vast range of techniques exists for determining the
optimum according to the specific problem. LP is commonly used for problems exhibiting
linearity in their statements [Ringertz, 1985], otherwise quadratic and non-linear methods
must be used [Bekdaş et al., 2017]. Specific cases, such Integer Programming (IP), take
into account discrete integer values which can be desired in a size optimisation when
availability of the real structural profiles is reduced.
While a large number of available procedures rely on NLP approach showing good
results for typical cases [Sedaghati and Esmailzadeh, 2003], [Adeli and Kamal, 1986] or
[Arora and Haug, 1976] , some other proposed algorithms are based on the Optimality Cri-
teria (OC) as a shortcut to find optima in a limited way [Venkayya, 1978, Venkayya, 1971,
Fleury, 1979].
When applied to truss problems, the optimal solution is found iteratively with a struc-
tural reanalysis to account for changes in load distribution. The efficiency of OC methods
has been shown to be weakly dependent on the number of design variables holding an
advantage over mathematical programming techniques in that they are not restricted to
locally optimal solutions in the vicinity of the initial design. However, in structures with a
high degree of statical indeterminacy, changes in load distribution may mean the approach
still fails to locate the global optimum.
While some works prefer the use of the well known force method, most of the visited
literature prefers the use of a simple formulation based on the displaced method. The use
of one formulation or the other is largely a matter of taste and availability of a suitable
computer program [Przemieniecki, 1985].
One last visited methodology is the so-called Full Stressed Design (FSD).
The solution procedure given in [Li, 1990] proceeds by including a scaling factor, ξ
(equation 2.39), which is used as an indicator to individually identifies the current active
constraint governing the element’s cross section.
!
ujn σij −σin
ξ = max , , (2.39)
ūj [σ]i φi [σ]i
When the active constraint is the stress-related one, the cross section at the ith element
for the design at iteration n is going to be updated in the following manner:
" ! #
k+i σin
(ai ) =k+i max , kµi (k ai ) (2.40)
[σ]i
In the other possible case, when the active constraint is a displacement constraint the
update must be computed as follows:
(" n
! n
!#η )
V
k+i k+i
σPq σPj P /Ep
X X
(ai ) = l 1/ lp ai (2.41)
ūj P =1 P =1
where η is a relaxation parameter used to control the stability and convergence of the
method and commonly takes a value between 0.1 and 0.2 [Li, 1990].
The algorithm showing the solution procedure is summarised in the flux diagram showed
in figure 2.7.
Strictly speaking, the FSD is not really an optimisation method but an automatic
technique of design. This technique is applicable to design for strength only and cannot
deal with more general types of constraints such as displacement-based constraints.
The FSD is principally used for statically determinate design problems in which strength
considerations govern over stiffness, nevertheless its use is proved to solve also statically
indeterminate structures within few analyses [Razani, 1965]. For highly redundant struc-
tures, due to the large number of possible fully stressed designs, the FSD algorithm may
diverge from the solution or oscillate about the optimum. Latter difficulty can be avoided
by restricting the evolution of the cross section to a determined percentage of their current
value [Vanderplaats, 1984b].
As seen in figure 2.7, the algorithm increases the size of over-stressed members and
reduces the size of under-stressed ones, reanalysing and iterating until the convergence is
achieved. However, Mueller [Mueller and Burns, 2001] demonstrates that this procedure
may exclude a set of feasible designs in which some members will respond to an increase
in size by attracting greater stress.
The first complication derives from the fact that the element cross section, ai , could
approach or even reach a zero value, which has obvious repercussion on the diagonal of
the stiffness matrix. To overcome this, the possibility of zero cross sections is no longer
permitted and in most cases, an inferior limit, amin , is imposed for ai (ai > 0 or ai ≥ amin ).
A non-zero lower bound will generally produce "secondary" elements whose only purpose
is often only to guaranty the non-singularity condition on the stiffness matrix and to avoid
inner mechanisms on the structure. Such elements are often erased or simply ignored at the
last stage of the optimisation, [Ohsaki and Swan, 2002]. This decision implies that most
Starting design 0 ai
ujk
σij −σik
ξ = max ,
[σ]i φi [σ]i
ξ = max ūj
Active
constraint
k
W −k−1 W
kW ≤
no
yes
Stop
Figure 2.7: Flowchart design procedure using FSD (inspired from [Li, 1990]).
optimal designs have a singular matrix and present potential mechanisms when described
as a part of the ground structure leading to the second listed complication.
The third complication is related to the choice of the ground structure. The ground
structure approach may or may not lead to the optimal structure according to the group of
nodes proposed (quantity and position) and the set of allowed elements; the optimal struc-
ture appears to be limited by the original geometrical restrictions and possible connections
(figure 2.8).
Figure 2.8: ST models obtained from 3 different ground structures [Gaynor et al., 2012].
more of the considered materials present different properties for negative and positive
stresses.
Figure 2.9 depicts a material where the absolute yield stress σc and σt (stresses of
compression and traction) are different, in the same manner Ec and Et may be different
[Achtziger, 1996]. This "equivalent" behaviour considers the case where two materials are
present in a structure. One branch of the stress-strain relation represent the behaviour of
the bars in compression while the other concerns the elements in compression.
A key point of this kind of optimisation must be the correct handling of the property-
assignation scheme. However, the nature of the forces at each element is not available at
the beginning of the process. Thus, this difficulty must be stated on the formulation of the
problem.
2.4 Summary
Most optimal design studies applied on skeletal structures deal with cross-sectional design
variables. However, due to the weight reductions that can be gained by modifications of
the structural configuration [Gil and Andreu, 2001], a growing quantity of scientific papers
interested in shape optimisation have been published in recent years. Yet, there are some
basic difficulties involved in this type of optimisation. One major complication is related to
the need of declare the existence or absence of structural elements. Another complication,
and probably the most important, involves the choice of the starting selection of nodes and
elements; the selection of the ground structure.
So far, the procedure to select a ground structure seems to relay on the designer experi-
ence. Intuitively, if the nodal positions are fixed, the choice of a saturated ground structure
would lead to the best results. In this manner, the geometric optimisation is "replaced"
by a large quantity of potential (fixed) nodes in the ground structure. The selection of
the "best nodal coordinates" is indirectly made by vanishing those nodes only attached to
elements whose cross section approximates to 0. From an opposed point of view, if nodal
positions are also implicitly included in the variable vector, a sparse ground structure is
preferred.
From a practical point of view, optimisation formulations that simultaneously improve
topology and geometry may lead to reasonable structures presenting the disadvantage of
being a highly non-linear problem and, in some cases, with disjointed design spaces. On the
other hand, pure ground structure optimisation generally its simple performed. However,
it has been shown that such approach frequently leads to singular unstable structures or
to structures with a considerable high quantity of secondary elements.
In an attempt to obtain reasonable structures through a simple algorithm, avoiding
the complications related with mixed design vectors, some methodologies are presented as
2-part optimisation problems.
The optimum solution can be obtained by the application of a stress-ratio method over
a fixed geometry followed by a geometric optimisation. In the first instance, this process
will allow to "reduce" the structure by decreasing the section of non-trivial elements. Once
the sizing procedure is completed, the optimisation of the geometry is implemented.
The proposed procedure combines the ground structure approach with a subsequent
iteration procedure that alternates between the member sizing optimisation (cross sections
as variables) and the shape modification (node coordinates as variables).
In addition, [Vanderplaats, 1984a] developed a technique where the stress-ratio method
was used to size the structure while keeping the topology fixed and the steepest descent
method was used to move the nodal coordinates while keeping the sections fixed. Using two
separate design spaces reduces the design variables at each sub-problem and simplifying
the global optimisation. The main drawback was that the algorithm was not able to
automatically to add or delete members or joints during the design process frequently
producing ill-conditioning problems.
According to the reviewed literature, a suitable way to threat truss optimisation problems
would be an hybrid algorithm able to sequentially threat the structure and a subroutine
capable of delete or insert elements without compromising the stability of the Stiffness
matrix and the whole structure.
In recent years, several approaches have been proposed in order to ease the application of
the Strut-and-Tie method. While some methodologies threat the problem as a continuous
optimisation, some others prefer the ground structure and truss optimisation to create an
acceptable reinforcement layout.
One of the main difficulties of this last approach, based on the truss optimisation, is
how to define the initial structure: the initial number of elements, connectivity and, the
most important, how to determine the number and position of the nodes.
The objective of this chapter is to present in detail the proposed computer aided
process to acquire Strut-and-Tie models for the reinforcement bi-dimensional and three-
dimensional concrete structures.
This chapter contains two important sections. The first one presents and describes the
algorithm of the proposed method. The second section intends to clear out some points of
the behaviour and the performance of the process seen on the test campaign.
• to ease the generation of ST models based on the linear-elastic stress field distribution
• to propose both a fully automatic approach and a rational one based on the experience
of the engineer
• to present a fast and low cost (in terms of computational effort) option to assess the
need of steel reinforcement respecting the EuroCodes
Despite the previously discussed limitations that a ST based on its linear elastic stress
field may have. It was decided that, according to some sources such as [Chae and Yun, 2016],
[Schlaich et al., 1987] and the [British Standards Institution, 2005], a strut-and-tie model
can be developed for a given element based on its linear elastic (uncracked) stress field, used
to identify from it a possible resisting truss model. Accepting that plausible ST models
can be produced through the application of structural optimisation techniques, the project
proposed a scheme based on weight minimisation. Within this work, discrete representa-
tions (FE models including only bar-type elements) have been preferred over the so-called
continuous optimisation. This decision was principally based on the structural behaviour
of a ST system (according to the EuroCodes) and also, based on the advantages that this
type of representation have in terms of its simplicity of treatment and the direct extraction
of the final results.
Since the beginning, the construction of the initial ground structure was one of the
major difficulties to overcome. In order to automatise the process, the selection of the
nodes, as well as the construction of the ground structure, is based on the linear elastic
stress field of a FE analysis of the structure. This selection allows to generate an initial
truss structure and proceed to optimise it.
The algorithm has been divided in five subroutines that include the analysis and the
post-process of a model. The included subroutines proceed to fulfil different tasks:
3. construction of the ground structure based on the trajectories of the principal stresses
product of the previous subroutine
Input.
Mechanical properties,
mesh, boundary conditions.
Linear-elastic FE analysis.
Principal stresses and
associated directions.
Ground Structure.
Division of the geometry, initial
strut path, linking elements.
Iterative process
Truss optimisation.
• Size
• Topology
• Geometric
Update
k ∗ no
≤
yes
Strut-and-tie model.
Nodal revision, secondary struts,
steel projection to the global axis.
Figure 3.1: Flowchart showing proposed the design procedure
3.1.1 Input
The input required for the algorithm depends on whether an initial FE analysis is needed
or, the user already posses the results of an analysis carried out on a specialised software.
For the case where the initial finite element analysis is desired to be treated by the
algorithm, the input data matches the requirements of any other FE solver: the geometry
depicted by a nodal list and a connectivity matrix, the list of materials, and the bound-
ary conditions. On the other hand, the algorithm can post-treat the results obtained by
software such as ANSYS and Code-Aster. For this case, additionally to the previous lists,
the input must contain the resultant principal stresses computed at the element’s Gauss
points.
The subroutine reads the available information and stocks it as matrices and vectors in
Matlab environment and inside a file where further results are also saved.
% K_global .− G l o b a l s t i f f n e s s matrix . 4
% k_m .− Element s t i f f n e s s matrix . 5
% DoF .− D e g r e e s o f freedom v e c t o r [ True F a l s e ] . 6
% P .− Load v e c t o r . 7
% U .− G l o b a l d i s p l a c e m e n t v e c t o r . 8
% Geom .− C e l l c o n t a i n i n g n o d a l l i s t and c o n n e c t i v i t y matrix 9
% Mat .− M a t e r i a l p r o p e r t i e s 10
%% Required s u b r o u t i n e s 11
% e l e m _ s t i f f .− Computes t h e e l e m e n t s t i f f n e s s matrix 12
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 13
K_global = z e r o s (num(DoF) ) ; %P r e l o c a t i n g g l o b a l s t i f f n e s s matrix 14
U = z e r o s (num(DoF) , 1 ) ; %P r e l o c a t i n g d i s p l a c e m e n t v e c t o r 15
f o r m = 1 : num_of_elem 16
k_m = e l e m _ s t i f f ( Mat , Geom ,m) ; %Element s t i f f n e s s matrix 17
K_global=K_global+k_m ; 18
end ; %Assembling t h e matrix 19
U(DoF==True )=P(DoF==True ) /K(DoF==True , DoF==True ) ; %S o l v e f o r U 20
The results of an initial linear elastic finite element model constitute the base of the
proposed algorithm.
Figure 3.2: Polynomial interpolation over 8 random XY Z triplets (second degree function of two variables of
the form f (x, y) = α1 x2i + α2 xi yi + α3 yi2 + α4 xi + α5 yi + α6 ).
2) the principal direction of the minor stresses at each sub-domain must be developed along
a remarkable preponderant direction.
y4
A1
y3
x A2
A4 Original data y2
A3 Interpolated data Original data
Stolen area y1 Interpolated data
Voronoi Interpolation grid
tessellation
x1 x2 x3 x4
(a) (b)
Figure 3.3: Grid based numerical differentiation: (a) Natural neighbours [Boissonnat and Cazals, 2000] and,
(b) Grid interpolated data
Intuitively, the directional derivative can be computed by selecting one variable and
numerically differentiate the function keeping the other variables fixed. Selecting a list
of fixed values for one variable implies that the differentiation neglects those results
that are not strictly placed along the selected value which leads to two possibilities:
1) the differentiation is done along all the values of the variables or 2) just some
values are selected. To avoid this, the numerical differentiation is done over a grid of
data obtained by a natural neighbour interpolation [Sibson, 1981] of the form:
n
X
σ(x, y, z) = wi σ(xi , yi , zi ) (3.1)
i=1
where σ(x, y, z) is the estimate value of the stress at coordinates (x, y, z), σ(xi , yi , zi )
and wi are the data values and their associated weights located at (xi , yi , zi ). The
weights are calculated superposing two Voronoi tessellations (refer to section 3.1.3.2),
one of only the data coordinates and another one including both the data and the
interpolation coordinates, and computing the "stolen" area at each surrounding zone
[Ledoux and Gold, 2005].
After having performed the interpolation of the data to approximate the values at
the intersections of the grid, the local maximum/minimum are searched over parallel
planes along the and axis through a numerical partial derivatives.
The partial derivatives are sensed to give the slope of some function f at any point
(a,b) in the directions parallel to the coordinate axes. From the definition of partial
derivatives (equations 3.2) given by [Levy, 2010], it can be appreciated that the partial
derivative ∂f /∂x is obtained by fixing the value of y to a constant b and differentiating
the function f (x, b) at x = a. In the same manner, the partial derivative with respect
to y is obtained by fixing x = a and differentiating the function f (a, y) at y = b.
∂f f (a + h, b) − f (a, b)
(a, b) = lim (3.2a)
∂x h→0 h
∂f f (a, b + h) − f (a, b)
(a, b) = lim (3.2b)
∂y h→0 h
2. Each subdivision must be big enough to be able to contain one straight element
within its limits.
3. At the same time, it must be small enough to present a clearly preponderant direction
of σIII (considering only the results of the initial FE model located at the interior of
each division).
The Voronoi diagram, also known as Voronoi tessellation, is the partition of an space
into a finite number of regions based on the distance between points in a specific subset of
known data; the "seeds" xi (i = 1, ..., n). Each seed, xi , is surrounded by a convex polygon
V (xi ) delimiting a Voronoi cell, which is defined as a set of points x that are closer to xi
than to any other seed xj (j 6= i) (see equation 3.3).
inside a sub-domain, a small disturbance is introduced over the position of the seeds. This
disturbance is included by replacing the original selected seeds by two new seeds result
from the intersection of a circle, of an arbitrary infinitesimal radius and centred at the
original seed’s coordinates, with the boundaries of the geometry.
1 2
wi θs (X, Y ) − AT θIII,j
X
Fit A to minimise: = (3.6)
j=1
Additionally to the kernel function, the loess also requires a specification of neighbour-
hood size. The chosen neighbourhoods, Ωi , are selected as cuboids whose maximum interior
diagonal connects the minimum and the maximum of the coordinates of the current cell.
Placing the kernel function over a cuboid zone of dimensions lX , lY and lZ (sides
measured along X, Y and Z respectively) and centred at the coordinates (CX , CY and CZ )
the local weight of the data subsets within the zone Ωi is given, according to their spatial
positions on the geometry (X, Y and Z) by the product of the parabolic distributions.
Within this step, the subroutine implemented to approach the initial strut inclination
which is computed based on the direction of the major principal stresses of the elements
found in the subdomain through the loes. This approach is adopted for subdomains pre-
senting a small angular dispersion and a significant quantity of available data. The initial
strut at subdomains presenting either a large angular dispersion or a small quantity of data
(less than 10) is then associated to the mean value of the circular directions Θ̄.
n
X n
X
Θ̄ = arctan 2 sin Θj , cos Θj (3.10)
j=1 j=1
where Θj represents the angle of the principal direction of the n elements found inside
the subdomain.
calculated for the contouring zone. The projected line is cut at the limits of the current
zone generating an ending node Nf . The next iteration takes the previous final node as
the new initial one, and proceeds in the same manner to calculate another final node. The
procedure continues for all the seeds until the boundaries of the geometry have been reached
by the generated “branches”. Hence, the strut generation algorithm produces branches that
start at the coordinates of concentrated loads and compressive stress concentrations, and
diffuses throughout design domain. Even though two or more branches could pass through
common zones, until this step, these branches do not necessarily share common nodes.
Therefore, a nodal merging procedure has been implemented at this state.
Ground structure
Iterative process
Ku = P
Sizing optimisation
∗
update
update
do not
Stability control
∗
Strut-and-tie model
Figure 3.4: Flowchart of the proposed truss optimisation
In order to decrease highlight a ST model from the ground structure it was implemented
a optimisation scheme able to reduce the quantity of elements, to choose the material of
each member (steel or concrete) and most important, to assess cross sections.
3.1.4.1 Initialisation
An initial configuration built by a total of m elements can be now stated based on the
strut path and the tie distribution.
At the very early stages of the Thesis, the initial cross section of the elements was
intended to be based on an energetic equivalence between the initial FE model and the
ground structure. The equivalence was made by computing the strain energy associated to
{−}
the compressive stresses only, ΨEF , within the FE model and calibrate it with the energy
P {−}
of the deformation of the proposed struts ΨS .
considering that for a general body the strain energy to a compressive state of stress
can be computed as:
A {−}
{−} 2 {−} {−}
Ψ{−} = λ I + II + 2µ (I )2 + (II )2 (3.11)
2
Analogously, in the case of a straight element undergoing an axial compressive force,
the strain energy can be easily expressed in terms of the force, F, cross section, ai , length,
l, strain, , and its Young’s modulus, E as:
1
Ψ{−} = E({−} )2 (3.12)
2
Considering that the struts are proposed and delimited within an individual Voronoi
zone, Ωi , the equilibrium is made individually for each strut; the strain energy from the
FE considers only the nodal results from the n nodes contained inside the limits of the
current zone leading to:
n
A {−} 1
{−} 2 {−} {−}
+ 2µ (I )2 + (II )2 = Eal({−} )2S,Ωj
X
λ I + II (3.13)
j=1 2 Ωj 2
{−}
So far, the axial strain of the struts S,Ωj is unknown but, looking for the compatibility of
displacements between the FE model and the ST one, the strain of each strut is estimated
as the nodal displacement between the 2 nodes closest to the strut edges divided by their
initial Euclidean distance leaving the cross section, a, as the only unknown.
{−} {−} 2 {−} {−}
Pn
A
j=1 2 λ I + II + 2µ (I )2 + (II )2
Ωj
a= (3.14)
El ({−} )2S,Ωj
This attempt to predict the cross section of the struts was rapidly discarded for two
main reasons: 1) the operations needed involved three-dimensional nodal research which
requires high computational efforts and 2) the results were, for most of the cases, too far
from the final result which was translated into the need of more iterations or even the
non-convergence.
For instance, the guess value of cross sections is proposed as a constant vector of value
ainitial . Considering the fact that the selection of the initial cross sections may affect the
performance, or even the solution of the optimisation, ainitial may consider a realistic steel
distribution. The user should chose a value that produce elements that individually fit into
the boundaries of the geometry.
The initial model considers that all elements are made out of steel. This consideration
is only made once. During the iterative process, the selection of the material is done based
on the sign of the force obtained at previous iterations. The material models correspond
to linear-elastic hypothesis whose behaviour law is assumed with Young’s moduli, Ei , and
Poisson’s ratio, ν, assigned according to the idealised selected materials.
1/2
maximum root
Condition number = κ(A) = (3.15)
minimum root
Although there is not a specific limit on the condition number to determine the
singularity of a matrix [Pyzara et al., 2011], for practical purposes some restrictions
were imposed. The minimum value that a cross section may take at the k-th iteration
is equal to the value of the largest cross section, computed at iteration k−1 divided by
106 . This value has found to be significantly small to represent a neglectable element
but to be sufficiently large to avoid numerical problems providing an inexpensive
damping to the inversion.
Ki = Ki−1 + ke i=i+1
ill- yes
conditioned
matrix
no
K = Ki
Figure 3.5: Element suppression flowchart
is superior to the minimum, and 2) the second phase randomly adds the rest of
the elements verifying that the stability of the stiffness matrix. The second phase
continues until the algorithm determines that no ill-condition is presented in the
problem and eliminates the reminder elements . The process is achieved through
the modification to the assembling subroutine (see flowchart 3.5) by adding a logical
operation that verifies the section of the current element and the conditioning of the
matrix once the current element has been added.
Even though suppressing all the elements presenting active constraints have little
repercussion on the conditioning of a stiffness matrix associated to a fully populated
structure, for a problem with sparsity and bandedness, erasing elements becomes a
trivial task. The presented subroutine guaranties the invertibility of the matrix and
erases non-trivial elements reducing the quantity of elements in the final structure
and also reducing the need of computational effort for further iterations.
To overcome the first complication logical restrictions were imposed. The algorithm
limits the final position of the nodes accepting the optimised coordinates if and only
if they are included within the geometry. For cases where the optimum coordinates
are found outside the geometry, the algorithm retains the computed direction α and
scales the step S to fit the structure. This allows keeping a classic formulation for
a geometric optimisation and, at the same time, it allows to directly include spatial
restrictions.
The second complication is linked with bar elements whose length is or approaches
to zero thus, their stiffness K = Ea/l approaches to the, computationally speaking,
infinite or is simply undefined.
According to [Achtziger, 2007], melting nodes can be frequently observed; in some
zones, the nodes tend to approach to the supports avoiding long elements with im-
portant stresses. Nevertheless, the melting nodes can cause ill-conditioning, their
presence can also be useful to control the number of elements in the structure.
From a practical point of view, one solution to threat this problem is to define
a minimum for the length of the elements. Another solution to work-around this
problem is to formally exclude the melting nodes from the structure.
For this work an alternative process was stated. An element presenting both, an
infinite stiffness and an infinitesimal length suggest that the initial and ending node
of such element may be merged into one single node. According to a distance specified
by the user or by the used precision (h), a subroutine selects the melting nodes and
evaluates if they can or cannot be merged. If the geometry considering the merge
leads to a stable structure, the suppression of the element proceeds and the model
is updated. Alternately, if the merge leads to an unstable structure at the current
iteration, the length of the element is increased.
3.1.5 Termination
The termination of the algorithm can occur in two different manners:
1. the current error, k attains the prescribed threshold i
2. the number of iterations has reached the prescribed maximum
C
C
C
C T
(a) (b)
Figure 3.6: Hydrostatic nodes: (a) CCC node geometry and (b) Tension force anchored by a plate
• Tension ties are anchored behind the node and considered struts.
• The area of the converted strut is equal to the area of concrete surrounding the
tension tie reinforcement and having the same centroid.
In other words, the tensile force induced by the ties is replaced by a compressive force
in the opposite face of the node. The resultant force is considered to act over a surface
computed under the same principle as the elements in compression: a = F/φσ. Consid-
ering that all the elements are conceived as prismatic geometries whose cross section was
computed considering the concrete’s allowable compressive strength, the resultant nodes
present are subjected to the same stress level at all their faces.
Based on the geometry obtained, a further verification of the bearing capacity of the
nodes can be approximated trough a linear elastic FE analysis. The principal stresses are
compared to the concrete’s allowable compressive strength using a Mohr-Coulomb failure
criterion.
3.1.6.2 Reinforcement
Considering the distribution of the steel reinforcement to follow the ties could result, from
a constructability point of view, in an impractical solution. In addition to present the
skeletal model, the algorithm proposes a solution that projects the ties to the principal
axis in order to obtain a more classical steel distribution.
A subproblem is stated for all considered ties where the oriented steel must replace the
existent element in the following fashion:
6
0.
A AR
W W
where ax and ay are the total steel areas needed along axis X and Y respectively; lx
and ly are the geometrical projections of the tie’s length over the reinforcement axis.
Equations 3.16 and 3.17 represent a simple optimisation problem that is solved within
the same subroutine.
(AR ∪ Ω) == Ω (3.18)
if the precedent statement is not true, the analysed element overpasses the feasible
section and an alert is printed
The nodes located at the geometric boundaries may be attached to elements that appear
to be slightly outside the permitted geometry. In reality, boundary nodes are shifted
towards the the interior of the geometry due to the concrete cover thickness.
Given that the algorithm does not consider a coverage thickness, most of the elements
located near the boundaries could be wrongly considered "outside" the geometry. To over-
come this, only a percentage of the total length of the element is matched with the feasible
section, Ω.
3.1.7 Outputs
The principal outputs of the algorithm are those associated with the quantity of reinforce-
ment, the position of the ties and nodes but also, the evolution of the optimisation process.
The outputs can be divided into three different groups.
The first type, and maybe the most important, groups the results of each iteration of
the geometry from the ground structure to the final ST model. Available on list format
(.txt) and on Visualization Toolkit (VTK), the results include the geometry (nodal list
and connectivity matrix); geometric characteristics of the elements such as length and
cross section; and the force developed at each bar. Additionally, data related to the nodal
revision is also included: geometry of the nodes and type of node (CCC, CTC, etc.). These
results are saved at each iteration into file named OUTPUT.
The second type corresponds to results that are stocked only in form of text such as
the reinforcement computed from the projection of the ties over the principal axis. If a
FE analysis was performed, the initial mesh, global displacements, stresses and principal
stresses can also be found in the OUTPUT file. Other intermediate results are also asso-
ciated to this group. Useful data of the iterative process such as the convergence rate and
the evolution of the total volume are also presented in text format.
The last type has been associated to results that are contained only in a non-interactive
visual representation. This information such as the Voronoi division, the subdomains
and the local maximum/mminimum of the principal stresses is only depicted in a Joint
Photographic Experts Group format (.jpeg). Although, the backup of this information
was not considered important, the data can be directly read and saved from the Matlab’s
interface.
3.2.1 Input
After having imposed the respective boundary conditions and solving some results were
extracted. The information extracted from ANSYS software was the nodal list (NLIST in
ANSYS environment), the connectivity matrix (ELIST), the principal stresses computed
at the Gauss points and also smoothed at the nodes (PRESOL and PRNSOL respectively)
and the direction field of the principal stresses (PVSOL). This information, schematised
in figures 3.9 and 3.10, is the starting point to develop the ground structure.
1
0.03
0.3
0.3
0.03
0.08
00.3
0.132838
0
-0 0
742
0.1
137
0.18
0.2
5
0.3 0
0.2
0.08
47
.552
6
0.3
4
0.3 0.6
2
0.
.0
0.
0.6
00.6
0.4
0.
0.
0.13
0
3 0.08 0.03
0
76 16
0.2 0.23
0. 0.8.9
0.71 00.8
0.9
0
0.03
-0.04
0.9
0.9
8
0.50.61 6
0.3
4
0.6
-0.0
0.52 7
7
0.1
0.08
0.3
0
0
0.47
0.42
8
3
0.3 0.1
0.0
0.37
0.32
0.8
0.8
0.32
0.03
0.28
0.18
3
3
0.1
0.2
0.3
8
0.28
0.0
0.23
0.08
0.7
0.7
0.23
3
0.1
0.13
0.6
0.6
0.18
0.18
0.18
0.5
0.5
0.4
0.4
0.3
0.3
0
0
0
0
0
0
0.2
0.2
-0.04
-0.04
-0.04
-0.04
0.1
0.1
-0.04
0.18
8
0.1
-0.04
while the application of topology and geometry optimisation is optional. In other words,
the algorithm gives the possibility to the user to apply a simply sizing optimisation or
mixed schemes; sizing-topology, sizing-geometry, or sizing-topology-geometric.
Figure 3.13 shows the results obtained through the application of 2 mixed schemes:
sizing-topology and sizing-topology-geometric. The principal difference between the results
obtained by the two different mixed schemes lies in the quantity of elements conforming
the final Strut-and-Tie model. More subtile differences are found in the nodal coordinates.
A mean distance of 5.8cm was found in the position of the nodes from a model to the
other.
(a) (b)
Figure 3.12: Ground structure development: (a) merged strut path and (b) proposed ground structure
(m=447).
(a) m=117 i=100 (b) m=117, i= 63 (c) m=24, i=74 (d) m=24, i=74
Figure 3.13: Anchorage Strut-and-Tie models obtained from a fine mesh. Figure (a) shows the results
obtained through an optimisation scheme considering only size and topology techniques; figure (c) shows the
results obtained through an optimisation scheme considering size, topology, and geometric techniques. Figures
(b) and (d) are graphical representations of the associated cross sections corresponding to (a) and (c) .
of the results in the finite element method; a problem referring to the contrast in the so-
lutions for different mesh sizes or democratisation, fact that can decrease the reliability
of the numerical results. Being based on the FEM, it is natural to expect certain mesh-
dependency on the proposed approach. In order to have an idea of how the choice of the
initial mesh can influence the final results, the same example was performed with different
initial meshes and the results are presented in figures 3.14 and 3.15. Similarly to the results
presented in the previous section, the figures show the results achieved by the application
of mixed schemes sizing-topology and sizing-topology-geometric.
Along with the mesh discretisation showed in 4.2, another two different meshes were
used for the purpose of comparison, as shown in figures 3.14a and 3.15a. Figures 3.13,
3.14 and 3.15 show the obtained ST schemes for the three different meshes, in blue dashed
lines represent members in compression and other continuous red lines represent members
in tension.
(a) 288 elements (b) m=84, i=46 (c) m=84, i=46 (d) m=21, i=74 (e) m=21, i=74
Figure 3.14: Anchorage Strut-and-Tie models obtained from a coarse mesh. Figure (a) shows the results
obtained through an optimisation scheme considering only size and topology techniques; figure (c) shows the
results obtained through an optimisation scheme considering size, topology, and geometric techniques. Figures
(b) and (d) are graphical representations of the associated cross sections corresponding to (a) and (c) .
As it can be observed, the results obtained from different initial meshes do not radically
differ from one model to another when treated with the full optimisation scheme (sizing-
topology-geometric). This can be explained by the fact that the initial FE mesh is used
only as mean to propose the initial truss system. If the algorithm is able to find similar
local maximum and minimum from the principal stress fields, the Voronoi division will
define the same zones for the initial system.
Although the number of elements may not be the same from one model to other, it
can be appreciated that the obtained differences are mainly found in secondary elements
remaining the principal elements as a constant for the three different systems. Additionally
it can be pointed out that the position of the nodes present a good consistency in the
results specially for size-topology-geometric schemes (subfigures (e)). Even thought the
final nodal coordinates are not exactly the same, the largest discrepancy between models
has been found to be about 15 centimetres which, arguably, does not impact a structure
whose smallest dimension is equal to 3 meters.
(a) 723 elements (b) m=130, i=66 (c) m=130, i=66 (d) m=28, i=90 (e) m=28, i=90
Figure 3.15: Anchorage Strut-and-Tie models obtained from a locally refined mesh. Figure (b) shows the results
obtained through an optimisation scheme considering only size and topology techniques; figure (c) shows the
results obtained through an optimisation scheme considering size, topology, and geometric techniques. Figures
(b) and (d) are graphical representations of the associated cross sections corresponding to (a) and (c) .
In general, the obtained ST models show consistently the same characteristics of load-
transfer mechanisms, such as the force-spreading and transverse tensile stresses. Despite
being computed from different initial meshes, figures 3.13e, 3.14e and, 3.15e present huge
similarities in terms of quantity , distribution and position of principal elements. Small
differences remarkable in the quantity of "secondary" elements; elements that are present
only for stability issues or ill-conditioning in the stiffness matrix. The presence of this
elements, having a neglectable cross section area, do not affect the behaviour of the whole
structure. Therefore, a conclusion can be safely drawn that mesh sizes have little influence
on the final model and results.
1. Force-spreading. The force spreading mechanism shows consistency with the model
presented by Schlaich; for the case of the model presented by Zhong, the difference
is due to a discretisation of the load (2 point load instead 1 punctual one).
2. Burst deep. The three showed topologies present close values for the parameter
dburst 1.46, 1.35, and, 1.41 meters for model (a), (b) and, (c) respectively. With
P P
P
(-)
dburst
Q
(+)
(a) [Zhong et al., 2016] (b) [Schlaich et al., 1987] (c) Computed model
Figure 3.16: Different Strut-and-Tie model propositions
respect to the angle Θ, the proposed model presents a value of 37 degrees; inclination
10o greater than the presented by the cited authors (26o reported by Zhong and 28o
by Schlaich).
3. Transverse tensile elements. Beyond the burst deep is where the most remark-
able differences arise. Whether the literature models show just two straight struts,
the computed model propose a continuity of the of the load-transfer mechanism pre-
senting short elements following a natural spreading of the force.
The differences presented in the first and second point of the previous list refer to small
variations that, given the size of the analysed element, are acceptable in actual structures.
For the third point, the literature results propose long strut elements (more than 4 meters
long) that, according to the EuroCodes, may need further detailing an even secondary
strut and ties.
P P P
(a) [Zhong et al., 2016] (b) [Aashto L.R.F.D., 1998] (c) Computed model
Figure 3.17: Comparison of different Strut-and-Tie model propositions(anchorage e=0.75m)
satisfied with the final results, it was decided to let the user the liberty to add, erase or
modify the proposed Voronoi seed hence adapting a computer aided scheme over a fully
automatic one.
3.4 Summary
As it was pointed out, the proposed methodology is based on an organised list of subroutines
that allow to automatically obtain feasible ST models from common linear-elastic FE
analysis. The initial FE planar analysis intends to serve as a link between the structural
engineer, used to this type of analysis, and the rational approach ST.
The performed analyses show that:
1. the proposed methodology keeps the "spirit" of the manual ST models being obtained
through linear-elastic stress fields
2. the obtained results approach to the geometries obtained by different methods and
authors
3. the use of different mesh sizes have little influence on the final model and results
So far the results clearly show the capabilities for finding feasible ST systems but the
gains in terms of reinforcement still remain unsaid. Next chapter addresses to this issue
and compares usual practice techniques to compute the reinforcement of a D-region.
Comparative example
In previous chapter, the proposed approach was described and a simple bi-dimensional
literature case was used to compare the obtained results.
Even though the automatic generation of feasible planar Strut-and-Tie Models can be
advantageous, a wide variety of structures present geometries or boundary conditions that
can limit the application of planar models. Besides, due to the graphical limitations, clas-
sical reinforcement schemes based on ST models remain applicable only to considered 2D
structures where the effects in a third direction are just omitted. When facing 3D prob-
lems, within the industrial context, the geometries are commonly reduced to mere surface
representations based on shell or plate type elements and the need of steel reinforcement is
directly computed though an algorithm based on generalised forces such as Capra-Maury.
Specifically, within the nuclear civil works, industrial constraints have imposed an im-
moderate use of plate and shell formulations. The size of the buildings, its structural
behaviour as well as the need of efficient auditable methods, are just some aspects that
have encouraged this practice.
This chapter is intended to compare results from the industrial practice and those
possibly achieved through the use of the rational approach STM while applied on local
zones; pointing out the advantages of the implementation of detailing stages during the
design procedure. To this purpose a literature example treated from three points of view:
It is worth to mention that in no case the computation time will be a subject of study.
This decision has been taken under two premises: 1) the main purpose is to evaluate
the viability of the proposed algorithm and, 2) the code was written in MATLAB that,
arguably, may induced a slow execution [Aruoba and Fernández-Villaverde, 2015].
99
Exemple comparative
Dans le chapitre précédent, l’approche proposée a été décrite et un simple cas de littérature
bidimensionnelle a été utilisé pour comparer les résultats obtenus.
Même si la génération automatique de modèles planaires Strut-and-Tie réalisables peut
être avantageuse, une grande variété de structures présentent des géométries ou des con-
ditions aux limites qui peuvent limiter l’application de modèles planaires. En outre, en
raison des limitations graphiques, les schémas de renforcement classiques basés sur les
modèles ST restent applicables uniquement aux structures 2D considérées où les effets dans
une troisième direction sont simplement omis. Face aux problèmes 3D, dans le contexte
industriel, les géométries sont souvent réduites à de simples représentations surfaciques
basées sur des éléments de type coque ou plaque et le besoin de ferraillage est directement
calculé par un algorithme basé sur des forces généralisées comme Capra-Maury.
Plus précisément, dans les travaux de génie civil nucléaire, les contraintes industrielles
ont imposé une utilisation immodérée des formulations de plaques et de coques. La taille des
bâtiments, son comportement structurel ainsi que le besoin de méthodes auditables efficaces
ne sont que quelques aspects qui ont encouragé cette pratique.
Ce chapitre vise à comparer les résultats de la pratique industrielle et ceux éventuelle-
ment obtenus grâce à l’utilisation d’une approche rationnelle appliquée aux zones locales;
souligner les avantages de la mise en œuvre des étapes de détail au cours de la procédure
de conception. A cet effet, un exemple de littérature traité de trois points de vue:
Il est important de mentionner que le temps de calcul ne sera en aucun cas un sujet
d’étude. Cette décision a été prise sous deux prémisses: 1) le but principal est d’évaluer la
viabilité de l’algorithme proposé et, 2) le code a été écrit dans MATLAB qui, sans doute,
peut induire une exécution lente [Aruoba and Fernández-Villaverde, 2015].
1.00 m
0.15 m
0.70 m
1.00 m
0.50 m 0.60 m
In figures 4.2b and 4.2c are shown the principal stress fields all over the structural ele-
ment developed under the specified boundary conditions and in subfigure 4.2f the detected
peaks and valleys are depicted.
(a) FE mesh (708 elements) (b) σI direction field (c) σIII direction field
0
0.3
2
0
0.0
0.16
0.05
0
1
1
0.16
0.9
0.9
0.16
0.02
0.8
0.8
0.05
0
0.7
0.7
0
6
0.1
0.3 0.44
-0.05
0
0.86
0.72
0.3
0
0.1 0.02
0
0.02
0.58
0
0.44
0.6
0.6
0.48
6
0.16
67
0
6
0.3
0.1
-0.05
0.2
0.3
6
0.05
0.1
0.05
0.16
0.02
0.16
0.02
0.5
0.5
0
0.16
0.05
0.05
0.05
0
0.02
0.05
0.02
0.16
0.4
0.4
0
0.16
0.05
0.02
-0.12
0.02
0.16
0
0.02
0.37
0.26
0.3
0.3
0
0
5
0.0
0
0.16
0.02
0.2
0.2
0.1
0.1
0.02
0.05
0.05
0
0
0
0.02
0
(d) σI (Normed isolines) (e) σIII (Normed isolines) (f) Voronoi seeds
Figure 4.2: Schematic representation of the needed data
Taking the local maxima and minima as seeds for the Voronoi division and the geometry
as the feasible region, the division is performed. For this case, 20 cells where found during
the performed division 4.3a.
Following the presented methodology, the loess was applied at the centre of every cell.
Considering the principal directions of the all the elements found in the cell, the regression
was applied and the results are schematised in figure 4.3b.
The initial strut path (subfigure 4.3c) was achieved through the sub routine described
in the previous chapter. An initial node is selected (node containing a maxima of σIII )
Based on the prior strut path, linking elements are placed and the ground structure is
proposed.
For the presented example, the hybrid size-topology-geometry optimisation scheme was
preferred and the final geometry is shown in figure 4.5a.
T13
T10
T14
T12
T9
T5 T8
T7 T4
T16 T6
T3
T1
T2
The nodal revision consist in a graphical projection of the element considering their ac-
tual cross section. While elements in compression (struts) are directly considered, elements
in tension are replaced by an equivalent strut arriving at the opposite face of the node.
Being all elements designed at full stress, and being the geometry of the nodes directly
computed from the intersection of the elements, the stress attained at nodal zones equals
the concrete’s strength design.
Figure 4.6 depicts the geometry of a CCT node produced by the intersection of three
elements of the final ST model.
Strut
Tie
Intersection S6 T4
Nodal zone
Node vertex
Cross section
limit
nt
le T4
va
qui of:
E rce
fo S2
As it can be appreciated, the proposed model (4.5a) present a good geometry in terms
of repetition of members and nodal position compared to the optimal design achieved by
the SESO approach (4.5b). Both models, also present strong similarities with the model
proposed by Schlaich and similar repetition of tensile elements.
Tables 4.2 and 4.3 contain the information of the ties for the model depicted in figure
4.5a. The first table presents the needed steel considering the Strut-and-Tie model as
a skeletal structure. The second output, table 4.3, corresponds to the results of a more
realistic model that considers the cross sections distributed along the principal axis X and
Y as it is usually done in the construction field. This last result is achieved through the
optimisation subproblem stated in section 3.1.6.2.
Table 4.3: Steel reinforcement for the planar ST model: projection over the X and Y axis [cm3 ]
Force Required steel Length Total reinforcement
Tie
[KN] [cm2 ] [cm] [cm3 ]
X 43.3 0.96 19 18.28
T1
Y 100.9 2.24 44 98.66
X 58.7 1.3 28 36.52
T2
Y 67.8 1.51 33 49.72
X 50.9 1.13 32 36.2
T3
Y 3.33 0.07 21 1.55
X 6.1 0.14 45 6.1
T4
Y 5 0.11 37 4.11
X 281.5 6.26 45 281.5
T5
Y 249.7 5.55 40 221.96
X 15.5 0.34 2 0.69
T6
Y 151.4 3.36 21 70.65
X 116 2.58 47 121.16
T7
Y 117.4 2.61 48 125.23
X 60.1 1.34 48 64.11
T8
Y 3.3 0.07 2 0.15
X 52.5 1.17 50 58.33
T9
Y 0 0 0 0
X 29.5 0.66 9 5.9
T10
Y 107.3 2.38 36 85.84
X 58.2 1.29 25 32.33
T11
Y 143.3 3.18 61 194.25
X 166 3.69 24 88.53
T12
Y 254.8 5.66 38 215.16
X 108 2.4 32 76.8
T13
Y 96.4 2.14 29 62.12
X 6.3 0.14 2 0.28
T14
Y 127.9 2.84 45 127.9
X 125.6 2.79 15 41.87
T15
Y 212.2 4.72 25 117.89
X 16 0.36 2 0.71
T16
Y 162.4 3.61 22 79.4
Total 2323.9
For comparative purposes, in this and further examples, the considered geometry is
divided into four regions according to table 4.4: the arm, the forehead, the joint and the
foot.
Table 4.4: Steel reinforcement for the planar ST model: need of steel reinforcement per zone [cm3 ]
X reinforcement Y reinforcement Total
Arm 345.61 222.1 567.71
Forehead 274.88 803.17 1078.05
Joint 157.11 199.99 357.1
Foot 91.71 229.33 321.04
P P P
=869.31 =1454.59 =2323.9
Figure 4.7 shows the PI history attained by the structure along the the optimisation
process. The PI decreases with the removal of lowly stressed elements from the corbel at
the same time that the error measured between continuous iterations, , also is reduced.
1 20
error [%]
PI index
T15
T11
T13
0.5 10
T10
T14
T12
T9
T5 T8
T7 T4
T16 T6
T3
T1
T2
0 0
0 5 10 15 20 25 30 35 40
Number of iteration
With the same aim as in the previous examples, two variants of the mesh discretisation
were used to perform the same analysis (same materials and boundary conditions) and the
results are plotted in figures 4.8a and 4.8b.
(a) Coarse (184 elements) (b) 29 bars at iteration 84 (c) 19 bars at iteration 63
Figure 4.8: Coarse mesh
(a) Coarse (184 elements) (b) 35 bars at iteration 31 (c) 19 bars at iteration 190
Figure 4.9: Free mesh
In general, the differences found in the results of the application of the algorithm
over different meshes do not drastically modify the final ST model. Hence, as in the
previous chapter, the statement that the obtained ST models show consistently the same
characteristics of load-transfer mechanisms regardless the initial FE mesh remains true.
1. open software
In order to be consistent with the industrial practice, it has been chosen to model the
corbel with a heavily coarse mesh constructed by elements of nearly 0.5 meters per side.
This choice has been guided by the will of modelling a simple object exhibiting a typical
D-region discretisation found in structural junction. With that in mind, the presented
model can be seen as a local extract of bigger structural representation.
The shell-element model was built, calculated and then post-processed by means of
Code_aster software. The software includes a function able to calculate reinforcement
densities in shell and plates elements as a function of the elements of reduction: the
generalised forces.
The result (referred as FERRAILLAGE in the data structure) can be accessed in text form
or directly plotted within SALOME_MECA interface.
At the Ultimate Limit States that is considered, the software operator CALC_FERRAILLAGE
only calculates the bending reinforcement areas. The shear reinforcement has been hand
calculated and was added to the calculated values. The field of reinforcement that is
obtained through these two operations per each element is then:
• a longitudinal reinforcement density in the X direction of the element for the lower
face of the element (DNSXI);
• a longitudinal reinforcement density in the Y direction of the element for the lower
face of the element (DNSYI);
• a longitudinal reinforcement density in the Y direction of the element for the lower
face of the element (DNSYI);
DNSYS
DNSXI Y
DNSYI
DNSXI
X Neutral axis
Top
Bottom
d
hea
re
Fo
t m
in
Jo Ar
ot
Fo
Following the previous representation, a model was constructed using 14 shell elements
of constant thickness (0.50m and 0.70m, respectively, for the elements conforming the
column and the arm). The load of 0.5MN was considered distributed along the deep of
the element. Finally, fixed supports were considered at the bottom and at the top of the
geometry. A graphical representation of the model can be seen in subfigure 4.12a.
#10-4 #10-5
Distributed load 6
4
4
3
3
2
2
Z
Y
1 1
0 0
(a) Mesh (16 elements) (b) Top layer along X (c) Bottom layer along X
#10-4 #10-4 #10 -4
9
3.5
7
8
3
6
7
Required steel reinforcement [m2/m]
2.5
5 6
2 5
4
4
1.5
3
1
2
2
1 0.5
1
0 0 0
(d) Top layer along Y (e) Bottom layer along Y (f) Shear
Figure 4.12: Capra-Maury computed required steel reinforcement (Coarse mesh).
The quantities of computed steel reinforcement were distributed into 4 zones and the
results are presented in the next table.
Table 4.5: Computed need of steel reinforced per zone (coarse model) [cm3 ]
XS XI YS YI SW Total
Arm 55.11 0 0 3555.56 154 3764.67
Forehead 0 98 239.22 607.11 0 944.33
Joint 35 230.22 29.44 508 125.22 927.89
Foot 17.81 111.11 0 0 0 128.92
Total 107.92 439.33 268.67 4670.67 279.22 5765.81
Precedent results are attained via a model representation that intends to mimic a
common mesh discretisation found in overall structural models. Assuming no singularities
are present, meshes dictated by overall models present an acceptable accuracy and might
be costly efficient for global behaviour objectives, yet their use for local zones and detailing
is highly questionable.
Following the good practices, in this context, a coarse model can be used for pre-design
purposes and after, local zones must be isolated and refined in order to proceed with the
detailing stage. Nevertheless, the coarse representation would be kept.
Worthy to mention is the fact that even if the selected discretisation may probably not
be the most appropriated for the problem, due to the size of nuclear facilities and their
correspondent buildings, the FE mesh herein used represents a common case industrially
used. Further, the selection of Finite Element representation may not be the best option
neither for local zones.
In order to decide when to use shell elements, there exist different thumb rules pre-
sented by several authors; [Johnson, 1986] presents three criteria to justify the use of shell
elements:
The selected corbel presents surface area / thickness ratios of 1.21 and 2.7 respectively
for the arm and the column. These values remain far from the application of shell elements
(being 10 the minimum acceptable value for surface area / thickness ratio [Johnson, 1986]).
Additionally, due to the geometry, shell elements can "overlap" at corners affecting the
precision of the results. Stiffness is still quite good, but stress at joints is not accurate even
for thin elements [Moaveni, 2011].
4.4 Remarks
Due to its relative ease of implementation, most of the structures are analysed and designed
in similar fashion as the one presented. The advantages of the use of shell elements to
model structural elements, results mainly from time-saving due to reduced number of
finite elements (and consequently the equations to solve). However, the shell elements are
limited to thinner features. There is not an specific limit of the application of surface
elements but a good approximate would be the ratio span / thickness [Bisch, 2013]. If the
ratio is at least 20 times, the bodies are candidates for surface elements otherwise, the use
of this type of elements need further justifications.
In an oversimplified definition, a shell addresses to a solid that presents a dimension
considerably smaller than the other two dimensions. Physically, and according to the
theory of structures, this fundamental characteristic allows the allusion to the hypotheses
that entail the simplification of the behaviour of three-dimensional bodies into the one of
a 2D body. Seldom, when modelling local zones, questionable simplifications are done.
Shell elements are used to model entire geometries indiscriminately. Whether it results
reasonable for static analysis of planar elements such as slabs or walls, its implementation
for regions presenting weak span/thickness ratios or geometric discontinuities is debatable.
For structural elements where the third direction cannot be neglected, solid element
discretisations are preferred.
In order to increase the effects of the load along the third direction, it was decided
to modify the thickness of the original model according to figure 4.14. The corbel was
assumed to present a one-meter thickness instead of the original 0.3m. Regarding the load,
an equivalent linearly distributed pattern was preferred over then punctual one.
0
1.0
P
1.00
1.00
0.15
1.00
P 0.1
5
0.70
0.70
0.70
1.00
1.00
0.6
0
1.00
0.5
0
A linear-elastic model consisting of 4200 8-node brick elements was built (figure 4.15a),
the boundary conditions and the load were applied and, finally, the stresses were found.
With the associated results, the algorithm was run and the obtained results are shown.
Figures 4.15 and 4.16 show some intermediate results product of the different steeps
that conform the algorithm.
(c) 86 Clipped
(a) Mesh (b) Seeds Voronoi division
Figure 4.15: Caption
(a) 3D view (b) Side view (c) Front view (d) Top view
Figure 4.17: Filtered Strut and tie model resultant of a size-topology scheme (m = 88)
Figure 4.17 shows the "filtered" ST model obtained through a size-topology optimisation
scheme. This image represents the actual sizes and distribution of the elements conforming
the proposed ST model. As in can be appreciated despite of having performed the filter,
there are still few elements whose existence may be not necessary for the structure. Hence,
in order to obtain a simpler ST model, a manual intervention may be a good option at this
state.
Another alternative is to apply the full optimisation scheme (size-topology-geometry).
Figures 4.18 and 1.17 show the results obtained through the application of the full
optimisation scheme (size-topology-geometry) to the ground structure previously presented
(figure 4.16d) .
Again, but this time after 329 iterations, the output shows a highly populated structure.
The scheme improved the results obtaining a final structure containing only 247 elements
(representing a 12% of the initial quantity). Despite the fact that the second optimisation
scheme allows a huge reduction regarding the quantity of elements of the initial structure,
the results are still far from being directly presented as a suitable ST model. At this point,
a minor manual intervention would be the ideal way to reveal the final ST model.
The proposed manual intervention is carried out by modifying the position of some
nodal coordinates. For this case the priority was given to those nodes generating in-
consistency in the symmetry along Z axis. From this point, after having performed the
modification of nodal coordinates, the optimisation scheme was retaken. The process con-
tinued for 120 iterations (additionally to the 329 already performed) and ended with a final
structure formed of only 86 elements (figure 4.19). The connectivity matrix and the nodal
list are included in the annex C.
(a) 3D view (b) Side view (c) Front view (d) Top view
Figure 4.19: Manually modified Strut and tie model resultant from a size-topology-geometry scheme (m = 86)
Table 4.6: Computed need of steel reinforced per zone (3D ST) [cm3 ]
Zone Total
Arm 927.3
Forehead 1086.83
Joint 382.33
Foot 198.63
Total 2595.1
4.6 Results
The following table presents the total need of steel reinforcement per zone for the four differ-
ent models presented: the surface representations where the steel reinforcement was com-
puted using the Capra-Maury algorithm, the planar and the solid representations treated
through the computer aided ST approach.
Table 4.7: Comparative table of need of steel reinforcement per zone [cm3 ]
Zone CM (coarse) ST (2D) ST (3D)
Arm 3764.67 567.72 927.3
Forehead 944.33 1078.05 1086.83
Joint 927.89 357.1 382.33
Foot 128.92 321.04 198.63
Total 5765.81 2323.9 2595.1
Table 4.8: Comparative table of need of steel reinforcement per zone. kg of steel per cubic meter of concrete
Zone Volume [m3 ] CM (coarse) ST (2D) ST (3D)
Arm 0.42 70.36 10.61 17.33
Forehead 0.50 14.83 16.93 17.06
Joint 0.35 20.81 8.01 8.58
Foot 0.50 2.02 5.04 3.12
P
=1.77 X̄ =27.01 X̄ =10.15 X̄ =11.52
∗
The density of steel was taken as 7.85 g/cm3
As it can be inferred, the quantities herein presented, are raw results and should be
arranged and smoothed before going to the blueprints. These arrangements may slightly
increase the shown reinforcement densities.
4.7 Discussion
From the results obtained several conclusions can be done. The comments will be divided
into two topics:
• Rational-based approach. All the sub-routines were inspired in the practice of the
engineer and remain close to the steep followed during a fully manual ST approach.
From the examples exposed, we note that the algorithm is able to identify feasible ST
models from a self created ground structures. As it can be appreciated, the worked exam-
ples tend to present a good behaviour for planar structures however, some complications
have been encountered in 3D cases:
• Time consumption. Even though it has been said that computation time would
not be considered, a comment should be made for the application of size-topology-
geometry optimisation scheme in 3D structures. While the feasible values and re-
strictions for cross sections remain unchanged when passing from a bi-dimensional
problem to a three-dimensional one, the number of plausible nodal positions and
restrictions increase. The time consumption is widely affected by specially three-
dimensional Boolean operations needed to verify if the a plausible nodal modification
remains inside the feasible domain (structure).
Review of contributions
Chapter 1 presents a summary of the current practice engineering design addressed to
reinforcement computation. In this chapter, efforts were made to point out the benefits
of rational approaches, specifically the strut-and-tie approach, to threat local members.
These techniques present the advantage of allowing a designer to follow the forces through
a structure with discontinuities (either static or geometric) which formerly were beyond
the scope of engineering practice based on the flexural theory. So far, one of the main
drawbacks of its application is the need of highly skilled structural engineers for a manual
approach or, the need of important resources in terms of computational effort for an,
arguably, less appropriate continuous optimisation.
Within Chapter 2, the ground structure approach arises as a suitable option to extract
feasible ST models from solid structures. The inconveniences of the ground structure
lie in the fact that a low populated ground structure itself needs to be proposed based
on experience leading again to the need of highly skilled structural engineers. Another
possibility is to initially propose high populated ground structures and introduce powerful
optimisation algorithms.
Chapter 3 proposes an alternative algorithm able to automatically propose and design
ground structures from the results of bi- or three-dimensional linear elastic analysis.
• Strut pattern search. The fashion developed to construct the initial strut pattern
structures was directly inspired from the recommendations made by Schlaich and also
the thumb rules found principally in the Eurocodes and the ACI. This aspect allowed
to keep the methodology as close as possible to the manual approach. The division of
the geometry based on the Voronoi tessellation, as well as the application of the loess
regression over the direction of the stress fields, are original ideas developed during
this work.
• Construction of ground structures. Even though the use of ground structures
has been applied before to find ST models, the definition of such structures is prede-
fined by the user and automatically developed as in the approach proposed.
• Optimised ST models. The selected optimisation scheme is presented as an ad-
vantageous tool for weight (cost) reduction in the engineering domain. Despite of
the benefits that the inclusion of optimisation techniques may have in the industrial
field, its application remain mostly for research purposes. This project examines the
application of this type of techniques to cases commonly encountered in the building
industry.
122
Referring to the objectives, this work attained to merge the rational approach known as
Strut-and-Tie with an automatic optimisation. The central hypothesis was confirmed given
that the proposed methodology and the implemented algorithm allow to obtain suitable
ground structures based on the direction fields of principal stresses. The results show that
it is also plausible to automatically obtain a optimum ST pattern from the proposed ground
structure.
Future work
This work represents a first stage of a potential future project that intends to develop an
open-source analysis tool.
Given the results attained within these three years, the author presents the tasks where,
according to the experience, future work should focus.
• 3D cases. As it was stated, the algorithmic advances attained in this work are
perfectible and can be optimised to reduce the manual intervention. Even if the pro-
cedure can lead to ST models, for three-dimensional cases, the implemented algorithm
presents minor drawbacks that were not encountered for bi-dimensional structures.
• Application in practice. Being based on a full stress design, the proposed algo-
rithm suits a serviceability limit state. Further modifications may have be done in
order to be used for ultimate limit state.
Projected trends
In recent years it can be observed a trend to develop ST models for steel concrete. From
the point of view of the mechanics, a truss-like reinforcement would be preferred over the
traditional grids. For decades the use of steel reinforcement has been aligned to principal
constructive axis mainly due to technical (economical) constraints. With the development
of industrial 3D printings this may change. Potential advantages of this process include
faster construction, lower labour costs and the production of less waste.
The reinforcement of real structures according to strut-and-tie models can represent
an suitable alternative. Truss-like reinforcement following stress trajectories would help to
reduce the existent gap between the linear-elastic models (commonly used for the design)
and the real structures.
Closing notes
In summary, it has been shown that this research has satisfied the objective of reducing the
evident gap between the rational approach known as Strut-and-Tie and automatic applied
methodologies based on finite element analysis industrially applied.
The algorithm developed within this work can be seen as one more alternatives that the
structural engineer has to safely design specific zones of the reinforced concrete structures.
Even if some algorithmic challenges, specially in terms of process optimisation, are still
to be addressed, the proposed methodology could guide the structural engineer through
the design of D-regions using strut-and-tie models.
[Achtziger, 1996] Achtziger, W. (1996). Truss topology optimization including bar proper-
ties different for tension and compression. Structural and Multidisciplinary Optimization,
12(1):63–74.
[ACI-318, 2008] ACI-318 (2008). Building code requirements for structural concrete : (aci
318-95) ; and commentary (aci 318r-95). American Concrete Institute.
[Adeli and Kamal, 1986] Adeli, H. and Kamal, O. (1986). Efficient optimization of space
trusses. Computers & structures, 24(3):501–511.
[Aguilar et al., 2002] Aguilar, G., Matamoros, A. B., Parra-Montesinos, G. J., Ramírez,
J. A., and Wight, J. K. (2002). Experimental evaluation of design procedures for shear
strength of deep reinfoced concrete beams.
[Almeida et al., 2013] Almeida, V. S., Simonetti, H. L., and Neto, L. O. (2013). Compara-
tive analysis of strut-and-tie models using smooth evolutionary structural optimization.
Engineering Structures, 56:1665–1675.
[Alshegeir and Ramirez, 1992] Alshegeir, A. and Ramirez, J. (1992). Computer graphics
in detailing strut-tie models. Journal of computing in civil engineering, 6(2):220–232.
[Armer et al., 1968] Armer, G., Mills, H., and Wood, R. (1968). Reinforcement of slabs in
accordance with a pre-determinated field of moments. Concrete, 2(8):319.
[Arora and Haug, 1976] Arora, J. and Haug, E. (1976). Efficient optimal design of struc-
tures by generalized steepest descent programming. International Journal for Numerical
Methods in Engineering, 10(4):747–766.
[Baldock and Shea, 2006] Baldock, R. and Shea, K. (2006). Structural topology optimiza-
tion of braced steel frameworks using genetic programming. In Intelligent Computing in
Engineering and Architecture, pages 54–61. Springer.
[Barton et al., 1991] Barton, D., Anderson, R., Bouadi, A., Jirsa, J., and Breen, J. (1991).
An investigation of strut-and-tie models for dapped beam details.
[Bathe, 2006] Bathe, K.-J. (2006). Finite element procedures. Klaus-Jurgen Bathe.
[Bazaraa et al., 2013] Bazaraa, M. S., Sherali, H. D., and Shetty, C. M. (2013). Nonlinear
programming: theory and algorithms. John Wiley & Sons.
[Bekdaş et al., 2017] Bekdaş, G., Nigdeli, S. M., and Türkakin, O. H. (2017). Non-linear
programing for sizing optimization of truss structures.
[Bendsøe et al., 1994] Bendsøe, M. P., Ben-Tal, A., and Zowe, J. (1994). Optimization
methods for truss geometry and topology design. Structural optimization, 7(3):141–159.
[Bendsøe and Kikuchi, 1988] Bendsøe, M. P. and Kikuchi, N. (1988). Generating optimal
topologies in structural design using a homogenization method. Computer methods in
applied mechanics and engineering, 71(2):197–224.
[Bendsøe and Sigmund, 2003] Bendsøe, M. P. and Sigmund, O. (2003). Topology optimiza-
tion: theory, methods and applications. 2003. Springer.
[Birrcher et al., 2009] Birrcher, D., Tuchscherer, R., Huizinga, M., Bayrak, O., Wood,
S. L., and Jirsa, J. O. (2009). Strength and serviceability design of reinforced concrete
deep beams. Technical report.
[Bisch, 2013] Bisch, P. (2013). Mécanique des coques: théorie et applications. Presses des
Ponts.
[Boissonnat and Cazals, 2000] Boissonnat, J.-D. and Cazals, F. (2000). Natural neighbour
coordinates of points on a surface. PhD thesis, INRIA.
[Capra and Maury, 1978] Capra, A. and Maury, J. (1978). Calcul automatique du ferrail-
lage optimal des plaques ou coques en béton armé. Annales de l’insitut technique du
bâtiment et des travaux publics, 367.
[Chae and Yun, 2016] Chae, H.-S. and Yun, Y. M. (2016). Strut-tie models and load
distribution ratios for reinforced concrete beams with shear span-to-effective depth ratio
of less than 3 (ii) validity evaluation. Journal of the Korea Concrete Institute, 28(3):267–
278.
[Fisher, 1995] Fisher, N. I. (1995). Statistical analysis of circular data. Cambridge Uni-
versity Press.
[for Structural Concrete, 2008] for Structural Concrete, I. F. (2008). Practitioners’ guide
to finite element modelling of reinforced concrete structures. State-of-Art Report.
[Foster et al., 2003] Foster, S. J., Marti, P., and Mojsilovic, N. (2003). Design of reinforced
concrete solids using stress analysis. Structural Journal, 100(6):758–764.
[Fraternali et al., 2011] Fraternali, F., Marino, A., Sayed, T. E., and Cioppa, A. D. (2011).
On the structural shape optimization through variational methods and evolutionary
algorithms. Mechanics of Advanced Materials and Structures, 18(4):225–243.
[Ganzreli, 2013] Ganzreli, S. (2013). Direct fully stressed design for displacement con-
straints. In 10th World Congress on Structural and Multidisciplinary Optimization, pages
19–24.
[Gaynor et al., 2012] Gaynor, A. T., Guest, J. K., and Moen, C. D. (2012). Reinforced
concrete force visualization and design using bilinear truss-continuum topology opti-
mization. Journal of Structural Engineering, 139(4):607–618.
[Gil and Andreu, 2001] Gil, L. and Andreu, A. (2001). Shape and cross-section optimisa-
tion of a truss structure. Computers & Structures, 79(7):681–689.
[Gordon, 2009] Gordon, J. E. (2009). Structures: or why things don’t fall down. Da Capo
Press.
[Grandić et al., 2015] Grandić, D., Šćulac, P., and Grandić, I. Š. (2015). Shear resistance
of reinforced concrete beams in dependence on concrete strength in compressive struts.
Technical Gazette, 22(4):925–934.
[Griffith and Stewart, 1961] Griffith, R. E. and Stewart, R. (1961). A nonlinear program-
ming technique for the optimization of continuous processing systems. Management
science, 7(4):379–392.
[Haftka and Gürdal, 2012] Haftka, R. T. and Gürdal, Z. (2012). Elements of structural
optimization, volume 11. Springer Science & Business Media.
[Hernández, 1993] Hernández, S. (1993). Del diseño convencional al diseño óptimo. posi-
bilidades y variantes. parte i. análisis de sensibilidad y optimización local y global.
[Herve et al., 2014] Herve, G., Cloitre, L., El Kadiri, S., Secourgeon, E, Mennela, D., and
Carême, C. (2014). Comparative study of shell element and brick element models for
npp structures. In Proceedings of TINCE 2014 Conference, pages –.
[Hofer and McCabe, 1998] Hofer, A. A. and McCabe, S. L. (1998). Comparison of shear
capacity of t-beams using strut and tie analysis. Technical report, University of Kansas
Center for Research, Inc.
[Hsu, 1992] Hsu, T. T. (1992). Unified theory of concrete concrete, volume 5. CRC press.
[Johansen, 1962] Johansen, K. W. (1962). Yield-line theory. Cement and Concrete Asso-
ciation.
[Kennedy and Goodchild, 2004] Kennedy, G. and Goodchild, C. (2004). Practical yield
line design. The concrete centre, Riverside house, 4.
[Khot and B., 1979] Khot, N. S., B. L. and B., V. V. (1979). Comparison of optimality
criteria algorithms for minimum weight design of structures. AIAA Journal, 17(2):182–
190.
[Kirsch, 1989] Kirsch, U. (1989). Optimal topologies of truss structures. Computer Methods
in Applied Mechanics and Engineering, 72(1):15–28.
[Klee and Minty, 1970] Klee, V. and Minty, G. J. (1970). How good is the simplex algo-
rithm. Technical report, WASHINGTON UNIV SEATTLE DEPT OF MATHEMAT-
ICS.
[Kuchma and Tjhin, 2001] Kuchma, D. A. and Tjhin, T. N. (2001). Cast (computer aided
strut-and-tie) design tool. In Structures 2001: A Structural Engineering Odyssey, pages
1–7.
[Kvasnica et al., 2004] Kvasnica, M., Grieder, P., Baotic, M., and Morari, M. (2004). Multi-
parametric toolbox (mpt). In HSCC, pages 448–462. Springer.
[Ledoux and Gold, 2005] Ledoux, H. and Gold, C. (2005). An efficient natural neighbour
interpolation algorithm for geoscientific modelling. Developments in spatial data han-
dling, pages 97–108.
[Lew and Narov, 1983] Lew, I. P. and Narov, F. (1983). Three-dimensional equivalent
frame analysis of shearwalls. Concrete International, 5(10):25–30.
[Li, 1990] Li, X. (1990). Truss structure optimum design and its engineering application.
Computers & Structures, 36(3):567–573.
[Marti, 1978] Marti, P. (1978). Plastic analysis of reinforced concrete shear walls.
Birkhäuser.
[Marti, 1985] Marti, P. (1985). Basic tools of reinforced concrete beam design. In Journal
Proceedings, volume 82, pages 46–56.
[Marti, 1990] Marti, P. (1990). Design of concrete slabs for transverse shear. Structural
Journal, 87(2):180–190.
[Marti et al., 1987] Marti, P., Leesti, P., and Khalifa, W. U. (1987). Torsion tests on
reinforced concrete slab elements. Journal of Structural Engineering, 113(5):994–1010.
[Mattock et al., 1976] Mattock, A. H., Chen, K., and Soongswang, K. (1976). The behavior
of reinforced concrete corbels. PCI Journal, 21(2):52–77.
[May and Ganaba, 1988] May, I. and Ganaba, T. (1988). A full range analysis of reinforced
concrete slabs using finite elements. International Journal for Numerical Methods in
Engineering, 26(4):973–985.
[May and Lodi, 2005] May, I. and Lodi, S. H. (2005). Deficiencies of the normal moment
yield criterion for rc slabs. Proceedings of the Institution of Civil Engineers-Structures
and Buildings, 158(6):371–380.
[Moaveni, 2011] Moaveni, S. (2011). Finite element analysis theory and application with
ANSYS, 3/e. Pearson Education India.
[Mörsch, 1902] Mörsch, E. (1902). Der eisenbetonbau, seine anwendung und theorie. Wayss
and Freytag, AG, Im selbstverlag der Firma, Neustadt ad Haardt, May.
[Mueller and Burns, 2001] Mueller, K. M. and Burns, S. A. (2001). Fully stressed frame
structures unobtainable by conventional design methodology. International Journal for
Numerical Methods in Engineering, 52(12):1397–1409.
[Muttoni et al., 2015] Muttoni, A., Ruiz, M. F., and Niketic, F. (2015). Design versus as-
sessment of concrete structures using stress fields and strut-and-tie models. ACI Struc-
tural Journal, 112(5):605.
[Nagarajan et al., 2010] Nagarajan, P., Jayadeep, U., and Pillai, T. M. (2010). Application
of micro truss and strut and tie model for analysis and design of reinforced concrete
structural elements. Sonklanakarin Journal of Science and Technology, 31(6):647.
[Nielsen, 1964] Nielsen, M. P. (1964). Limit analysis of reinforced concrete slabs. Danish
Acad. of Technical Sciences.
[Nilson, 1997] Nilson, A. (1997). Design of concrete structures. Number 12th Edition.
[Ohsaki and Swan, 2002] Ohsaki, M. and Swan, C. (2002). Topology and geometry opti-
mization of trusses and frames. Recent advances in optimal structural design, 46.
[Park and Gamble, 2000] Park, R. and Gamble, W. L. (2000). Reinforced concrete slabs.
John Wiley & Sons.
[Park et al., 2006] Park, S. W., Linsen, L., Kreylos, O., Owens, J. D., and Hamann, B.
(2006). Discrete sibson interpolation. IEEE Transactions on Visualization and Computer
Graphics, 12(2):243–253.
[Pyzara et al., 2011] Pyzara, A., Bylina, B., and Bylina, J. (2011). The influence of a
matrix condition number on iterative methods’ convergence. In Computer Science and
Information Systems (FedCSIS), 2011 Federated Conference on, pages 459–464. IEEE.
[Ramirez and Breen, 1991] Ramirez, J. A. and Breen, J. E. (1991). Evaluation of a modi-
fied truss-model approach for beams in shear. Structural Journal, 88(5):562–572.
[Rao et al., 2007] Rao, G. A., Kunal, K., and Eligehausen, R. (2007). Shear strength of rc
deep beams. In Proceedings of the 6th International Conference on Fracture Mechanics
of Concrete and Concrete Structures, volume 2, pages 693–699.
[Razani, 1965] Razani, R. (1965). Behavior of fully stressed design of structures and its
relationship to minimum-weight design. AIAA Journal, 3(12):2262–2268.
[Ritter, 1899] Ritter, W. (1899). Die bauweise hennebique (the hennebique construction
method). Die Bauweise Hennebique, Schweizerische Bauzeitung, 33(7):41–61.
[Saniee, 2008] Saniee, K. (2008). A simple expression for multivariate lagrange interpola-
tion.
[Schittkowski et al., 1994] Schittkowski, K., Zillober, C., and Zotemantel, R. (1994). Nu-
merical comparison of nonlinear programming algorithms for structural optimization.
Structural and Multidisciplinary Optimization, 7(1):1–19.
[Schlaich et al., 1987] Schlaich, J., Schäfer, K., and Jennewein, M. (1987). Toward a con-
sistent design of structural concrete. PCI journal, 32(3):74–150.
[Schmith and Fox, 1965] Schmith, L. and Fox, R. (1965). An integrated approach to struc-
tural synthesis and analysis. AIAA JOURNAL, 3(6).
[Shah et al., 2011] Shah, A., Haq, E., and Khan, S. (2011). Analysis and design of dis-
turbed regions in concrete structures. Procedia Engineering, 14:3317–3324.
[Singh et al., 2018] Singh, B., Vimal, A., and Gaurav, G. (2018). Whither transverse
reinforcement in bottle-shaped struts? In Structures. Elsevier.
[Soltani and Corotis, 1988] Soltani, M. and Corotis, R. B. (1988). Failure cost design of
structural systems. Structural safety, 5(4):239–252.
[Starčev-Ćurčin et al., 2013] Starčev-Ćurčin, A., Rašeta, A., and Brujić, Z. (2013). Auto-
matic generation of planar rc strut-and-tie models. Facta universitatis-series: Architec-
ture and Civil Engineering, 11(1):1–12.
[Svanberg, 1981] Svanberg, K. (1981). On local and global minima in structural optimiza-
tion. Technical report, ROYAL INST OF TECH STOCKHOLM (SWEDEN).
[Swan et al., 1999] Swan, C. C., Arora, J. S., and Kocer, F. Y. (1999). Continuum and
ground structure topology methods for concept design of structures. In 1999 New Orleans
Structures Congress.
[Topkis and Veinott, 1967] Topkis, D. M. and Veinott, Jr, A. F. (1967). On the conver-
gence of some feasible direction algorithms for nonlinear programming. SIAM Journal
on Control, 5(2):268–279.
[Turmo et al., 2009] Turmo, J., Ramos, G., and Aparicio, A. (2009). Shear truss analogy
for concrete members of solid and hollow circular cross section. Engineering Structures,
31(2):455–465.
[Unnikrishna and Devdas, 2003] Unnikrishna, P. and Devdas, M. (2003). Reinforced con-
crete design.
[Van Kreveld et al., 2011] Van Kreveld, M., Van Lankveld, T., and Veltkamp, R. C.
(2011). On the shape of a set of points and lines in the plane. In Computer Graph-
ics Forum, volume 30, pages 1553–1562. Wiley Online Library.
[Vecchio and Selby, 1991] Vecchio, F. and Selby, R. (1991). Toward compression-field anal-
ysis of reinforced concrete solids. Journal of Structural Engineering, 117(6):1740–1758.
[Vecchio and Collins, 1986] Vecchio, F. J. and Collins, M. P. (1986). The modified
compression-field theory for reinforced concrete elements subjected to shear. In Journal
Proceedings, volume 83, pages 219–231.
[Venkayya, 1971] Venkayya, V. (1971). Design of optimum structures. Computers & Struc-
tures, 1(1-2):265–309.
[Venkayya, 1978] Venkayya, V. B. (1978). Structural optimization: a review and some rec-
ommendations. International Journal for Numerical Methods in Engineering, 13(2):203–
228.
[Wahlgren and Bailleul, 2016] Wahlgren, M. and Bailleul, N. (2016). Experimental analy-
sis of limitations in the strut-and-tie method.
[Williams et al., 2012] Williams, C., Deschenes, D., and Bayrak, O. (2012). Strut-and-tie
model design examples for bridges: Final report. Center for Transportation Research,
Report No. FHWA/TX-12/5-5253-01-1, Austin, Texas.
[Wood, 1961] Wood, R. H. (1961). Plastic and elastic design of slabs and plates: with
particular reference to reinforced concrete floor slabs. Thames and Hudson.
[Wright and Nocedal, 1999] Wright, S. J. and Nocedal, J. (1999). Numerical optimization.
Springer Science, 35(67-68):7.
[Yun and Lee, 2005] Yun, Y. and Lee, W. (2005). Nonlinear strut-tie model analysis of
pre-tensioned concrete deep beams. Advances in Structural Engineering, 8(1):85–98.
[Zhong et al., 2016] Zhong, J., Wang, L., Li, Y., and Zhou, M. (2016). A practical ap-
proach for generating the strut-and-tie models of anchorage zones. Journal of Bridge
Engineering, 22(4):04016134.
C.1 Manually modified Strut and tie model resultant from a size-topology-geometry
scheme (m = 86) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Other examples
P 1.00
0.25
1.00
0.25 P
(a) Voronoi division (b) Direction field of σI (c) Direction field of σIII
8 29 37
27
21
22 30
18 11 10 25
13 38 24
20
16 12 26
14
7 33
15 39
5
6 3
17 2 28
31
32
23
4
36 19 1
(d) Initial strut path (e) Merged strut path (f) Ground structure
Figure A.5: Steep of ST model
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
A planar representation of the structure was built up using a regular mesh consisting
in 1800 4-node plane stress elements (figure B.2a). The load was represented by a unique
punctual force. Concerning the support, all degrees of freedom were suppressed for the
nodes at the base zone. Plane strain hypothesis were considered during the procedure.
B.1.1 Input
After having imposed the respective boundary conditions and solving some results were
extracted. The information extracted from ANSYS software was the nodal list (NLIST in
143
ANSYS environment), the connectivity matrix (ELIST), the principal stresses computed
at the Gauss points and also smoothed at the nodes (PRESOL and PRNSOL respectively)
and the direction field of the principal stresses (PVSOL). This information, schematised
in figures B.2 and B.3, is the starting point to develop the ground structure.
1
0.14
0.1
0.05
4 14
0.91
0.1
0.81
0.76
0.05
0.05
0.71
0.67
0.1
0.05
0.62
0.20 0.1 .20.
0.57
0.52
0.71
38
0.1
0.05
0.2 0.0.19
0.62 7
14
0.57
30.4
0.52
9
0.6
0.48
9.3380
0.4
83
0.05
9
433
0.4
0.3
0.1
0.05
0.
0.0.29
0.24
0.9
0.9
0
0.1
0.14
0.19
0.1
0
0.05
0.14
0
0.14
0
0.8
0.8
0.1
0.05
0.1
0
0.0
0
0
0.1
0
0 0.05
0.7
0.7
0
0
0
0.6
0.6
0.1
0.05
0.5
0.5
0.05
0.4
0.4
0.05
0.3
0.3
0.1
0.2
0.2
5
0.0
0.1
0.1
0.05
0.05
5
0.0
0.1
0.05
Taking the local maximum and minimum as seeds for the Voronoi division and the
geometry as the feasible region, the division is performed. For this case, 20 cells where
found during the performed division (refer to subfigure B.5a).
Following the presented methodology, the loess was applied at the centre of every cell.
Considering the principal directions of the all the elements found in the cell, the regression
was applied and the results are schematised in figure B.5b.
The initial strut path (subfigure B.4d was achieved through the sub routine described
in the previous chapter. An initial node is selected (node containing a maximum of σIII )
and a straight line is developed until finding the limits of a cell).
(a) (b)
Figure B.5: Ground structure development: (a) merged strut path and (b) proposed ground structure (m=325).
(a) m=25 i=46 (b) m=25, i=46 (c) m=24, i=69 (d) m=24, i=69
Figure B.6: Anchorage Strut-and-Tie models obtained from a fine mesh. Figure (a) shows the results obtained
through an optimisation scheme considering only size and topology techniques; figure (c) shows the results
obtained through an optimisation scheme considering size, topology, and geometric techniques. Figures (b) and
(d) are graphical representations of the associated cross sections corresponding to (a) and (c) .
P P P
(a) [Zhong et al., 2016] (b) [Schlaich et al., 1987] (c) Computed model
Figure B.7: Comparison of different Strut-and-Tie model propositions
28
20 20 8
8 31
2.5
31 2.5
2.5
20
3 4
5 20 4
9
18 Y
18 19 13 6.2
2 31 16 14 2 28
2 2 2
27
10 X
14 16 2 18 3
30 30 27 6
16
7 9 12 21
1.5 10
30 1.5 17 7
11 6 1.5
26 5.8
17 12 6 19 4
1 17
1
12
29 1221 29 26 21
1 15 1 5.6 30 27
15 25 1 29 15 21
0.5
22 5.4
24
Z
0.5 24 25 21 6
0
0.5 24 Z
Y
23 Z 5.2
11 9
X 0
23 22 10
6 X
0
0.5 0 23 X
5
5.5 1 0.8 0.6 0.4 0.2 1
1 0.8
5 1.2 1 0.6 0.4 0.2 0
6.2 6 5.8 5.6 5.4 5.2 5
(a) 3D view (b) Side view (c) Front view (d) Top view
Figure C.1: Manually modified Strut and tie model resultant from a size-topology-geometry scheme (m = 86)
148